MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
/*
|
|
|
|
|
* HTTP compression.
|
|
|
|
|
*
|
|
|
|
|
* Copyright 2012 Exceliance, David Du Colombier <dducolombier@exceliance.fr>
|
|
|
|
|
* William Lallemand <wlallemand@exceliance.fr>
|
|
|
|
|
*
|
|
|
|
|
* This program is free software; you can redistribute it and/or
|
|
|
|
|
* modify it under the terms of the GNU General Public License
|
|
|
|
|
* as published by the Free Software Foundation; either version
|
|
|
|
|
* 2 of the License, or (at your option) any later version.
|
|
|
|
|
*
|
|
|
|
|
*/
|
|
|
|
|
|
|
|
|
|
#include <stdio.h>
|
2012-10-26 09:05:35 -04:00
|
|
|
|
MAJOR: compression: integrate support for libslz
This library is designed to emit a zlib-compatible stream with no
memory usage and to favor resource savings over compression ratio.
While zlib requires 256 kB of RAM per compression context (and can only
support 4000 connections per GB of RAM), the stateless compression
offered by libslz does not need to retain buffers between subsequent
calls. In theory this slightly reduces the compression ratio but in
practice it does not have that much of an effect since the zlib
window is limited to 32kB.
Libslz is available at :
http://git.1wt.eu/web?p=libslz.git
It was designed for web compression and provides a lot of savings
over zlib in haproxy. Here are the preliminary results on a single
core of a core2-quad 3.0 GHz in 32-bit for only 300 concurrent
sessions visiting the home page of www.haproxy.org (76 kB) with
the default 16kB buffers :
BW In BW Out BW Saved Ratio memory VSZ/RSS
zlib 237 Mbps 92 Mbps 145 Mbps 2.58 84M / 69M
slz 733 Mbps 380 Mbps 353 Mbps 1.93 5.9M / 4.2M
So while the compression ratio is lower, the bandwidth savings are
much more important due to the significantly lower compression cost
which allows to consume even more data from the servers. In the
example above, zlib became the bottleneck at 24% of the output
bandwidth. Also the difference in memory usage is obvious.
More tests run on a single core of a core i5-3320M, with 500 concurrent
users and the default 16kB buffers :
At 100% CPU (no limit) :
BW In BW Out BW Saved Ratio memory VSZ/RSS hits/s
zlib 480 Mbps 188 Mbps 292 Mbps 2.55 130M / 101M 744
slz 1700 Mbps 810 Mbps 890 Mbps 2.10 23.7M / 9.7M 2382
At 85% CPU (limited) :
BW In BW Out BW Saved Ratio memory VSZ/RSS hits/s
zlib 1240 Mbps 976 Mbps 264 Mbps 1.27 130M / 100M 1738
slz 1600 Mbps 976 Mbps 624 Mbps 1.64 23.7M / 9.7M 2210
The most important benefit really happens when the CPU usage is
limited by "maxcompcpuusage" or the BW limited by "maxcomprate" :
in order to preserve resources, haproxy throttles the compression
ratio until usage is within limits. Since slz is much cheaper, the
average compression ratio is much higher and the input bandwidth
is quite higher for one Gbps output.
Other tests made with some reference files :
BW In BW Out BW Saved Ratio hits/s
daniels.html zlib 1320 Mbps 163 Mbps 1157 Mbps 8.10 1925
slz 3600 Mbps 580 Mbps 3020 Mbps 6.20 5300
tv.com/listing zlib 980 Mbps 124 Mbps 856 Mbps 7.90 310
slz 3300 Mbps 553 Mbps 2747 Mbps 5.97 1100
jquery.min.js zlib 430 Mbps 180 Mbps 250 Mbps 2.39 547
slz 1470 Mbps 764 Mbps 706 Mbps 1.92 1815
bootstrap.min.css zlib 790 Mbps 165 Mbps 625 Mbps 4.79 777
slz 2450 Mbps 650 Mbps 1800 Mbps 3.77 2400
So on top of saving a lot of memory, slz is constantly 2.5-3.5 times
faster than zlib and results in providing more savings for a fixed CPU
usage. For links smaller than 100 Mbps, zlib still provides a better
compression ratio, at the expense of a much higher CPU usage.
Larger input files provide slightly higher bandwidth for both libs, at
the expense of a bit more memory usage for zlib (it converges to 256kB
per connection).
2015-03-28 21:32:06 -04:00
|
|
|
#if defined(USE_SLZ)
|
|
|
|
|
#include <slz.h>
|
|
|
|
|
#elif defined(USE_ZLIB)
|
2012-10-26 09:05:35 -04:00
|
|
|
/* Note: the crappy zlib and openssl libs both define the "free_func" type.
|
|
|
|
|
* That's a very clever idea to use such a generic name in general purpose
|
|
|
|
|
* libraries, really... The zlib one is easier to redefine than openssl's,
|
|
|
|
|
* so let's only fix this one.
|
|
|
|
|
*/
|
|
|
|
|
#define free_func zlib_free_func
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
#include <zlib.h>
|
2012-10-26 09:05:35 -04:00
|
|
|
#undef free_func
|
2012-10-31 06:19:18 -04:00
|
|
|
#endif /* USE_ZLIB */
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
|
2016-12-22 13:46:17 -05:00
|
|
|
#include <common/cfgparse.h>
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
#include <common/compat.h>
|
2017-07-25 05:07:15 -04:00
|
|
|
#include <common/hathreads.h>
|
2018-11-25 13:14:37 -05:00
|
|
|
#include <common/initcall.h>
|
|
|
|
|
#include <common/memory.h>
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
|
|
|
|
|
#include <types/global.h>
|
|
|
|
|
#include <types/compression.h>
|
|
|
|
|
|
2013-04-20 11:33:20 -04:00
|
|
|
#include <proto/acl.h>
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
#include <proto/compression.h>
|
2012-11-09 11:05:39 -05:00
|
|
|
#include <proto/freq_ctr.h>
|
2015-04-03 09:40:56 -04:00
|
|
|
#include <proto/stream.h>
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
|
2012-10-30 09:30:39 -04:00
|
|
|
|
2018-11-26 08:44:03 -05:00
|
|
|
#if defined(USE_ZLIB)
|
2018-11-25 14:12:18 -05:00
|
|
|
__decl_spinlock(comp_pool_lock);
|
2017-11-07 05:57:54 -05:00
|
|
|
#endif
|
2017-07-25 05:07:15 -04:00
|
|
|
|
2012-10-30 09:30:39 -04:00
|
|
|
#ifdef USE_ZLIB
|
|
|
|
|
|
2012-11-16 12:06:41 -05:00
|
|
|
static void *alloc_zlib(void *opaque, unsigned int items, unsigned int size);
|
|
|
|
|
static void free_zlib(void *opaque, void *ptr);
|
|
|
|
|
|
2012-10-30 09:30:39 -04:00
|
|
|
/* zlib allocation */
|
|
|
|
|
static struct pool_head *zlib_pool_deflate_state = NULL;
|
|
|
|
|
static struct pool_head *zlib_pool_window = NULL;
|
|
|
|
|
static struct pool_head *zlib_pool_prev = NULL;
|
|
|
|
|
static struct pool_head *zlib_pool_head = NULL;
|
|
|
|
|
static struct pool_head *zlib_pool_pending_buf = NULL;
|
|
|
|
|
|
2012-11-20 05:25:20 -05:00
|
|
|
long zlib_used_memory = 0;
|
2012-11-07 10:12:57 -05:00
|
|
|
|
2016-12-22 13:46:17 -05:00
|
|
|
static int global_tune_zlibmemlevel = 8; /* zlib memlevel */
|
|
|
|
|
static int global_tune_zlibwindowsize = MAX_WBITS; /* zlib window size */
|
|
|
|
|
|
2012-10-30 09:30:39 -04:00
|
|
|
#endif
|
|
|
|
|
|
2012-11-20 11:01:01 -05:00
|
|
|
unsigned int compress_min_idle = 0;
|
2012-11-16 12:06:41 -05:00
|
|
|
|
2015-03-28 10:46:00 -04:00
|
|
|
static int identity_init(struct comp_ctx **comp_ctx, int level);
|
|
|
|
|
static int identity_add_data(struct comp_ctx *comp_ctx, const char *in_data, int in_len, struct buffer *out);
|
2015-03-28 14:17:31 -04:00
|
|
|
static int identity_flush(struct comp_ctx *comp_ctx, struct buffer *out);
|
|
|
|
|
static int identity_finish(struct comp_ctx *comp_ctx, struct buffer *out);
|
2015-03-28 10:46:00 -04:00
|
|
|
static int identity_end(struct comp_ctx **comp_ctx);
|
|
|
|
|
|
MAJOR: compression: integrate support for libslz
This library is designed to emit a zlib-compatible stream with no
memory usage and to favor resource savings over compression ratio.
While zlib requires 256 kB of RAM per compression context (and can only
support 4000 connections per GB of RAM), the stateless compression
offered by libslz does not need to retain buffers between subsequent
calls. In theory this slightly reduces the compression ratio but in
practice it does not have that much of an effect since the zlib
window is limited to 32kB.
Libslz is available at :
http://git.1wt.eu/web?p=libslz.git
It was designed for web compression and provides a lot of savings
over zlib in haproxy. Here are the preliminary results on a single
core of a core2-quad 3.0 GHz in 32-bit for only 300 concurrent
sessions visiting the home page of www.haproxy.org (76 kB) with
the default 16kB buffers :
BW In BW Out BW Saved Ratio memory VSZ/RSS
zlib 237 Mbps 92 Mbps 145 Mbps 2.58 84M / 69M
slz 733 Mbps 380 Mbps 353 Mbps 1.93 5.9M / 4.2M
So while the compression ratio is lower, the bandwidth savings are
much more important due to the significantly lower compression cost
which allows to consume even more data from the servers. In the
example above, zlib became the bottleneck at 24% of the output
bandwidth. Also the difference in memory usage is obvious.
More tests run on a single core of a core i5-3320M, with 500 concurrent
users and the default 16kB buffers :
At 100% CPU (no limit) :
BW In BW Out BW Saved Ratio memory VSZ/RSS hits/s
zlib 480 Mbps 188 Mbps 292 Mbps 2.55 130M / 101M 744
slz 1700 Mbps 810 Mbps 890 Mbps 2.10 23.7M / 9.7M 2382
At 85% CPU (limited) :
BW In BW Out BW Saved Ratio memory VSZ/RSS hits/s
zlib 1240 Mbps 976 Mbps 264 Mbps 1.27 130M / 100M 1738
slz 1600 Mbps 976 Mbps 624 Mbps 1.64 23.7M / 9.7M 2210
The most important benefit really happens when the CPU usage is
limited by "maxcompcpuusage" or the BW limited by "maxcomprate" :
in order to preserve resources, haproxy throttles the compression
ratio until usage is within limits. Since slz is much cheaper, the
average compression ratio is much higher and the input bandwidth
is quite higher for one Gbps output.
Other tests made with some reference files :
BW In BW Out BW Saved Ratio hits/s
daniels.html zlib 1320 Mbps 163 Mbps 1157 Mbps 8.10 1925
slz 3600 Mbps 580 Mbps 3020 Mbps 6.20 5300
tv.com/listing zlib 980 Mbps 124 Mbps 856 Mbps 7.90 310
slz 3300 Mbps 553 Mbps 2747 Mbps 5.97 1100
jquery.min.js zlib 430 Mbps 180 Mbps 250 Mbps 2.39 547
slz 1470 Mbps 764 Mbps 706 Mbps 1.92 1815
bootstrap.min.css zlib 790 Mbps 165 Mbps 625 Mbps 4.79 777
slz 2450 Mbps 650 Mbps 1800 Mbps 3.77 2400
So on top of saving a lot of memory, slz is constantly 2.5-3.5 times
faster than zlib and results in providing more savings for a fixed CPU
usage. For links smaller than 100 Mbps, zlib still provides a better
compression ratio, at the expense of a much higher CPU usage.
Larger input files provide slightly higher bandwidth for both libs, at
the expense of a bit more memory usage for zlib (it converges to 256kB
per connection).
2015-03-28 21:32:06 -04:00
|
|
|
#if defined(USE_SLZ)
|
|
|
|
|
|
|
|
|
|
static int rfc1950_init(struct comp_ctx **comp_ctx, int level);
|
|
|
|
|
static int rfc1951_init(struct comp_ctx **comp_ctx, int level);
|
|
|
|
|
static int rfc1952_init(struct comp_ctx **comp_ctx, int level);
|
|
|
|
|
static int rfc195x_add_data(struct comp_ctx *comp_ctx, const char *in_data, int in_len, struct buffer *out);
|
|
|
|
|
static int rfc195x_flush(struct comp_ctx *comp_ctx, struct buffer *out);
|
|
|
|
|
static int rfc195x_finish(struct comp_ctx *comp_ctx, struct buffer *out);
|
|
|
|
|
static int rfc195x_end(struct comp_ctx **comp_ctx);
|
|
|
|
|
|
|
|
|
|
#elif defined(USE_ZLIB)
|
2015-03-28 17:08:25 -04:00
|
|
|
|
2015-03-28 10:46:00 -04:00
|
|
|
static int gzip_init(struct comp_ctx **comp_ctx, int level);
|
2015-03-28 12:00:39 -04:00
|
|
|
static int raw_def_init(struct comp_ctx **comp_ctx, int level);
|
2015-03-28 10:46:00 -04:00
|
|
|
static int deflate_init(struct comp_ctx **comp_ctx, int level);
|
|
|
|
|
static int deflate_add_data(struct comp_ctx *comp_ctx, const char *in_data, int in_len, struct buffer *out);
|
2015-03-28 14:17:31 -04:00
|
|
|
static int deflate_flush(struct comp_ctx *comp_ctx, struct buffer *out);
|
|
|
|
|
static int deflate_finish(struct comp_ctx *comp_ctx, struct buffer *out);
|
2015-03-28 10:46:00 -04:00
|
|
|
static int deflate_end(struct comp_ctx **comp_ctx);
|
2015-03-28 17:08:25 -04:00
|
|
|
|
2015-03-28 10:46:00 -04:00
|
|
|
#endif /* USE_ZLIB */
|
|
|
|
|
|
2012-10-30 09:30:39 -04:00
|
|
|
|
2012-11-10 13:27:47 -05:00
|
|
|
const struct comp_algo comp_algos[] =
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
{
|
2015-03-28 17:08:25 -04:00
|
|
|
{ "identity", 8, "identity", 8, identity_init, identity_add_data, identity_flush, identity_finish, identity_end },
|
MAJOR: compression: integrate support for libslz
This library is designed to emit a zlib-compatible stream with no
memory usage and to favor resource savings over compression ratio.
While zlib requires 256 kB of RAM per compression context (and can only
support 4000 connections per GB of RAM), the stateless compression
offered by libslz does not need to retain buffers between subsequent
calls. In theory this slightly reduces the compression ratio but in
practice it does not have that much of an effect since the zlib
window is limited to 32kB.
Libslz is available at :
http://git.1wt.eu/web?p=libslz.git
It was designed for web compression and provides a lot of savings
over zlib in haproxy. Here are the preliminary results on a single
core of a core2-quad 3.0 GHz in 32-bit for only 300 concurrent
sessions visiting the home page of www.haproxy.org (76 kB) with
the default 16kB buffers :
BW In BW Out BW Saved Ratio memory VSZ/RSS
zlib 237 Mbps 92 Mbps 145 Mbps 2.58 84M / 69M
slz 733 Mbps 380 Mbps 353 Mbps 1.93 5.9M / 4.2M
So while the compression ratio is lower, the bandwidth savings are
much more important due to the significantly lower compression cost
which allows to consume even more data from the servers. In the
example above, zlib became the bottleneck at 24% of the output
bandwidth. Also the difference in memory usage is obvious.
More tests run on a single core of a core i5-3320M, with 500 concurrent
users and the default 16kB buffers :
At 100% CPU (no limit) :
BW In BW Out BW Saved Ratio memory VSZ/RSS hits/s
zlib 480 Mbps 188 Mbps 292 Mbps 2.55 130M / 101M 744
slz 1700 Mbps 810 Mbps 890 Mbps 2.10 23.7M / 9.7M 2382
At 85% CPU (limited) :
BW In BW Out BW Saved Ratio memory VSZ/RSS hits/s
zlib 1240 Mbps 976 Mbps 264 Mbps 1.27 130M / 100M 1738
slz 1600 Mbps 976 Mbps 624 Mbps 1.64 23.7M / 9.7M 2210
The most important benefit really happens when the CPU usage is
limited by "maxcompcpuusage" or the BW limited by "maxcomprate" :
in order to preserve resources, haproxy throttles the compression
ratio until usage is within limits. Since slz is much cheaper, the
average compression ratio is much higher and the input bandwidth
is quite higher for one Gbps output.
Other tests made with some reference files :
BW In BW Out BW Saved Ratio hits/s
daniels.html zlib 1320 Mbps 163 Mbps 1157 Mbps 8.10 1925
slz 3600 Mbps 580 Mbps 3020 Mbps 6.20 5300
tv.com/listing zlib 980 Mbps 124 Mbps 856 Mbps 7.90 310
slz 3300 Mbps 553 Mbps 2747 Mbps 5.97 1100
jquery.min.js zlib 430 Mbps 180 Mbps 250 Mbps 2.39 547
slz 1470 Mbps 764 Mbps 706 Mbps 1.92 1815
bootstrap.min.css zlib 790 Mbps 165 Mbps 625 Mbps 4.79 777
slz 2450 Mbps 650 Mbps 1800 Mbps 3.77 2400
So on top of saving a lot of memory, slz is constantly 2.5-3.5 times
faster than zlib and results in providing more savings for a fixed CPU
usage. For links smaller than 100 Mbps, zlib still provides a better
compression ratio, at the expense of a much higher CPU usage.
Larger input files provide slightly higher bandwidth for both libs, at
the expense of a bit more memory usage for zlib (it converges to 256kB
per connection).
2015-03-28 21:32:06 -04:00
|
|
|
#if defined(USE_SLZ)
|
|
|
|
|
{ "deflate", 7, "deflate", 7, rfc1950_init, rfc195x_add_data, rfc195x_flush, rfc195x_finish, rfc195x_end },
|
|
|
|
|
{ "raw-deflate", 11, "deflate", 7, rfc1951_init, rfc195x_add_data, rfc195x_flush, rfc195x_finish, rfc195x_end },
|
|
|
|
|
{ "gzip", 4, "gzip", 4, rfc1952_init, rfc195x_add_data, rfc195x_flush, rfc195x_finish, rfc195x_end },
|
|
|
|
|
#elif defined(USE_ZLIB)
|
2015-03-28 17:08:25 -04:00
|
|
|
{ "deflate", 7, "deflate", 7, deflate_init, deflate_add_data, deflate_flush, deflate_finish, deflate_end },
|
|
|
|
|
{ "raw-deflate", 11, "deflate", 7, raw_def_init, deflate_add_data, deflate_flush, deflate_finish, deflate_end },
|
|
|
|
|
{ "gzip", 4, "gzip", 4, gzip_init, deflate_add_data, deflate_flush, deflate_finish, deflate_end },
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
#endif /* USE_ZLIB */
|
2015-03-28 11:40:46 -04:00
|
|
|
{ NULL, 0, NULL, 0, NULL , NULL, NULL, NULL, NULL }
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
};
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Add a content-type in the configuration
|
|
|
|
|
*/
|
|
|
|
|
int comp_append_type(struct comp *comp, const char *type)
|
|
|
|
|
{
|
|
|
|
|
struct comp_type *comp_type;
|
|
|
|
|
|
2016-04-03 07:48:43 -04:00
|
|
|
comp_type = calloc(1, sizeof(*comp_type));
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
comp_type->name_len = strlen(type);
|
|
|
|
|
comp_type->name = strdup(type);
|
|
|
|
|
comp_type->next = comp->types;
|
|
|
|
|
comp->types = comp_type;
|
|
|
|
|
return 0;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Add an algorithm in the configuration
|
|
|
|
|
*/
|
|
|
|
|
int comp_append_algo(struct comp *comp, const char *algo)
|
|
|
|
|
{
|
|
|
|
|
struct comp_algo *comp_algo;
|
|
|
|
|
int i;
|
|
|
|
|
|
2015-03-28 11:40:46 -04:00
|
|
|
for (i = 0; comp_algos[i].cfg_name; i++) {
|
|
|
|
|
if (!strcmp(algo, comp_algos[i].cfg_name)) {
|
2016-04-03 07:48:43 -04:00
|
|
|
comp_algo = calloc(1, sizeof(*comp_algo));
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
memmove(comp_algo, &comp_algos[i], sizeof(struct comp_algo));
|
|
|
|
|
comp_algo->next = comp->algos;
|
|
|
|
|
comp->algos = comp_algo;
|
|
|
|
|
return 0;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
return -1;
|
|
|
|
|
}
|
|
|
|
|
|
2016-08-10 15:17:06 -04:00
|
|
|
#if defined(USE_ZLIB) || defined(USE_SLZ)
|
2018-11-26 05:58:30 -05:00
|
|
|
DECLARE_STATIC_POOL(pool_comp_ctx, "comp_ctx", sizeof(struct comp_ctx));
|
|
|
|
|
|
2012-11-16 12:06:41 -05:00
|
|
|
/*
|
|
|
|
|
* Alloc the comp_ctx
|
|
|
|
|
*/
|
|
|
|
|
static inline int init_comp_ctx(struct comp_ctx **comp_ctx)
|
|
|
|
|
{
|
|
|
|
|
#ifdef USE_ZLIB
|
|
|
|
|
z_stream *strm;
|
|
|
|
|
|
2012-11-20 05:25:20 -05:00
|
|
|
if (global.maxzlibmem > 0 && (global.maxzlibmem - zlib_used_memory) < sizeof(struct comp_ctx))
|
2012-11-16 12:06:41 -05:00
|
|
|
return -1;
|
|
|
|
|
#endif
|
|
|
|
|
|
2017-11-24 11:34:44 -05:00
|
|
|
*comp_ctx = pool_alloc(pool_comp_ctx);
|
2012-11-16 12:06:41 -05:00
|
|
|
if (*comp_ctx == NULL)
|
|
|
|
|
return -1;
|
MAJOR: compression: integrate support for libslz
This library is designed to emit a zlib-compatible stream with no
memory usage and to favor resource savings over compression ratio.
While zlib requires 256 kB of RAM per compression context (and can only
support 4000 connections per GB of RAM), the stateless compression
offered by libslz does not need to retain buffers between subsequent
calls. In theory this slightly reduces the compression ratio but in
practice it does not have that much of an effect since the zlib
window is limited to 32kB.
Libslz is available at :
http://git.1wt.eu/web?p=libslz.git
It was designed for web compression and provides a lot of savings
over zlib in haproxy. Here are the preliminary results on a single
core of a core2-quad 3.0 GHz in 32-bit for only 300 concurrent
sessions visiting the home page of www.haproxy.org (76 kB) with
the default 16kB buffers :
BW In BW Out BW Saved Ratio memory VSZ/RSS
zlib 237 Mbps 92 Mbps 145 Mbps 2.58 84M / 69M
slz 733 Mbps 380 Mbps 353 Mbps 1.93 5.9M / 4.2M
So while the compression ratio is lower, the bandwidth savings are
much more important due to the significantly lower compression cost
which allows to consume even more data from the servers. In the
example above, zlib became the bottleneck at 24% of the output
bandwidth. Also the difference in memory usage is obvious.
More tests run on a single core of a core i5-3320M, with 500 concurrent
users and the default 16kB buffers :
At 100% CPU (no limit) :
BW In BW Out BW Saved Ratio memory VSZ/RSS hits/s
zlib 480 Mbps 188 Mbps 292 Mbps 2.55 130M / 101M 744
slz 1700 Mbps 810 Mbps 890 Mbps 2.10 23.7M / 9.7M 2382
At 85% CPU (limited) :
BW In BW Out BW Saved Ratio memory VSZ/RSS hits/s
zlib 1240 Mbps 976 Mbps 264 Mbps 1.27 130M / 100M 1738
slz 1600 Mbps 976 Mbps 624 Mbps 1.64 23.7M / 9.7M 2210
The most important benefit really happens when the CPU usage is
limited by "maxcompcpuusage" or the BW limited by "maxcomprate" :
in order to preserve resources, haproxy throttles the compression
ratio until usage is within limits. Since slz is much cheaper, the
average compression ratio is much higher and the input bandwidth
is quite higher for one Gbps output.
Other tests made with some reference files :
BW In BW Out BW Saved Ratio hits/s
daniels.html zlib 1320 Mbps 163 Mbps 1157 Mbps 8.10 1925
slz 3600 Mbps 580 Mbps 3020 Mbps 6.20 5300
tv.com/listing zlib 980 Mbps 124 Mbps 856 Mbps 7.90 310
slz 3300 Mbps 553 Mbps 2747 Mbps 5.97 1100
jquery.min.js zlib 430 Mbps 180 Mbps 250 Mbps 2.39 547
slz 1470 Mbps 764 Mbps 706 Mbps 1.92 1815
bootstrap.min.css zlib 790 Mbps 165 Mbps 625 Mbps 4.79 777
slz 2450 Mbps 650 Mbps 1800 Mbps 3.77 2400
So on top of saving a lot of memory, slz is constantly 2.5-3.5 times
faster than zlib and results in providing more savings for a fixed CPU
usage. For links smaller than 100 Mbps, zlib still provides a better
compression ratio, at the expense of a much higher CPU usage.
Larger input files provide slightly higher bandwidth for both libs, at
the expense of a bit more memory usage for zlib (it converges to 256kB
per connection).
2015-03-28 21:32:06 -04:00
|
|
|
#if defined(USE_SLZ)
|
|
|
|
|
(*comp_ctx)->direct_ptr = NULL;
|
|
|
|
|
(*comp_ctx)->direct_len = 0;
|
2018-07-10 11:43:27 -04:00
|
|
|
(*comp_ctx)->queued = BUF_NULL;
|
MAJOR: compression: integrate support for libslz
This library is designed to emit a zlib-compatible stream with no
memory usage and to favor resource savings over compression ratio.
While zlib requires 256 kB of RAM per compression context (and can only
support 4000 connections per GB of RAM), the stateless compression
offered by libslz does not need to retain buffers between subsequent
calls. In theory this slightly reduces the compression ratio but in
practice it does not have that much of an effect since the zlib
window is limited to 32kB.
Libslz is available at :
http://git.1wt.eu/web?p=libslz.git
It was designed for web compression and provides a lot of savings
over zlib in haproxy. Here are the preliminary results on a single
core of a core2-quad 3.0 GHz in 32-bit for only 300 concurrent
sessions visiting the home page of www.haproxy.org (76 kB) with
the default 16kB buffers :
BW In BW Out BW Saved Ratio memory VSZ/RSS
zlib 237 Mbps 92 Mbps 145 Mbps 2.58 84M / 69M
slz 733 Mbps 380 Mbps 353 Mbps 1.93 5.9M / 4.2M
So while the compression ratio is lower, the bandwidth savings are
much more important due to the significantly lower compression cost
which allows to consume even more data from the servers. In the
example above, zlib became the bottleneck at 24% of the output
bandwidth. Also the difference in memory usage is obvious.
More tests run on a single core of a core i5-3320M, with 500 concurrent
users and the default 16kB buffers :
At 100% CPU (no limit) :
BW In BW Out BW Saved Ratio memory VSZ/RSS hits/s
zlib 480 Mbps 188 Mbps 292 Mbps 2.55 130M / 101M 744
slz 1700 Mbps 810 Mbps 890 Mbps 2.10 23.7M / 9.7M 2382
At 85% CPU (limited) :
BW In BW Out BW Saved Ratio memory VSZ/RSS hits/s
zlib 1240 Mbps 976 Mbps 264 Mbps 1.27 130M / 100M 1738
slz 1600 Mbps 976 Mbps 624 Mbps 1.64 23.7M / 9.7M 2210
The most important benefit really happens when the CPU usage is
limited by "maxcompcpuusage" or the BW limited by "maxcomprate" :
in order to preserve resources, haproxy throttles the compression
ratio until usage is within limits. Since slz is much cheaper, the
average compression ratio is much higher and the input bandwidth
is quite higher for one Gbps output.
Other tests made with some reference files :
BW In BW Out BW Saved Ratio hits/s
daniels.html zlib 1320 Mbps 163 Mbps 1157 Mbps 8.10 1925
slz 3600 Mbps 580 Mbps 3020 Mbps 6.20 5300
tv.com/listing zlib 980 Mbps 124 Mbps 856 Mbps 7.90 310
slz 3300 Mbps 553 Mbps 2747 Mbps 5.97 1100
jquery.min.js zlib 430 Mbps 180 Mbps 250 Mbps 2.39 547
slz 1470 Mbps 764 Mbps 706 Mbps 1.92 1815
bootstrap.min.css zlib 790 Mbps 165 Mbps 625 Mbps 4.79 777
slz 2450 Mbps 650 Mbps 1800 Mbps 3.77 2400
So on top of saving a lot of memory, slz is constantly 2.5-3.5 times
faster than zlib and results in providing more savings for a fixed CPU
usage. For links smaller than 100 Mbps, zlib still provides a better
compression ratio, at the expense of a much higher CPU usage.
Larger input files provide slightly higher bandwidth for both libs, at
the expense of a bit more memory usage for zlib (it converges to 256kB
per connection).
2015-03-28 21:32:06 -04:00
|
|
|
#elif defined(USE_ZLIB)
|
2017-07-25 05:07:15 -04:00
|
|
|
HA_ATOMIC_ADD(&zlib_used_memory, sizeof(struct comp_ctx));
|
2012-11-16 12:06:41 -05:00
|
|
|
|
|
|
|
|
strm = &(*comp_ctx)->strm;
|
|
|
|
|
strm->zalloc = alloc_zlib;
|
|
|
|
|
strm->zfree = free_zlib;
|
|
|
|
|
strm->opaque = *comp_ctx;
|
|
|
|
|
#endif
|
|
|
|
|
return 0;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Dealloc the comp_ctx
|
|
|
|
|
*/
|
|
|
|
|
static inline int deinit_comp_ctx(struct comp_ctx **comp_ctx)
|
|
|
|
|
{
|
|
|
|
|
if (!*comp_ctx)
|
|
|
|
|
return 0;
|
|
|
|
|
|
2017-11-24 11:34:44 -05:00
|
|
|
pool_free(pool_comp_ctx, *comp_ctx);
|
2012-11-16 12:06:41 -05:00
|
|
|
*comp_ctx = NULL;
|
|
|
|
|
|
|
|
|
|
#ifdef USE_ZLIB
|
2017-07-25 05:07:15 -04:00
|
|
|
HA_ATOMIC_SUB(&zlib_used_memory, sizeof(struct comp_ctx));
|
2012-11-16 12:06:41 -05:00
|
|
|
#endif
|
|
|
|
|
return 0;
|
|
|
|
|
}
|
2016-08-10 15:17:06 -04:00
|
|
|
#endif
|
2012-11-16 12:06:41 -05:00
|
|
|
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
|
|
|
|
|
/****************************
|
|
|
|
|
**** Identity algorithm ****
|
|
|
|
|
****************************/
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Init the identity algorithm
|
|
|
|
|
*/
|
2015-03-28 10:46:00 -04:00
|
|
|
static int identity_init(struct comp_ctx **comp_ctx, int level)
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
{
|
|
|
|
|
return 0;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Process data
|
2012-11-19 06:35:37 -05:00
|
|
|
* Return size of consumed data or -1 on error
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
*/
|
2015-03-28 10:46:00 -04:00
|
|
|
static int identity_add_data(struct comp_ctx *comp_ctx, const char *in_data, int in_len, struct buffer *out)
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
{
|
2018-06-07 12:46:28 -04:00
|
|
|
char *out_data = b_tail(out);
|
2018-06-15 07:59:36 -04:00
|
|
|
int out_len = b_room(out);
|
2012-11-19 06:35:37 -05:00
|
|
|
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
if (out_len < in_len)
|
|
|
|
|
return -1;
|
|
|
|
|
|
|
|
|
|
memcpy(out_data, in_data, in_len);
|
|
|
|
|
|
2018-06-28 12:17:23 -04:00
|
|
|
b_add(out, in_len);
|
2012-11-19 06:35:37 -05:00
|
|
|
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
return in_len;
|
|
|
|
|
}
|
|
|
|
|
|
2015-03-28 14:17:31 -04:00
|
|
|
static int identity_flush(struct comp_ctx *comp_ctx, struct buffer *out)
|
|
|
|
|
{
|
|
|
|
|
return 0;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static int identity_finish(struct comp_ctx *comp_ctx, struct buffer *out)
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
{
|
|
|
|
|
return 0;
|
|
|
|
|
}
|
|
|
|
|
|
MAJOR: compression: integrate support for libslz
This library is designed to emit a zlib-compatible stream with no
memory usage and to favor resource savings over compression ratio.
While zlib requires 256 kB of RAM per compression context (and can only
support 4000 connections per GB of RAM), the stateless compression
offered by libslz does not need to retain buffers between subsequent
calls. In theory this slightly reduces the compression ratio but in
practice it does not have that much of an effect since the zlib
window is limited to 32kB.
Libslz is available at :
http://git.1wt.eu/web?p=libslz.git
It was designed for web compression and provides a lot of savings
over zlib in haproxy. Here are the preliminary results on a single
core of a core2-quad 3.0 GHz in 32-bit for only 300 concurrent
sessions visiting the home page of www.haproxy.org (76 kB) with
the default 16kB buffers :
BW In BW Out BW Saved Ratio memory VSZ/RSS
zlib 237 Mbps 92 Mbps 145 Mbps 2.58 84M / 69M
slz 733 Mbps 380 Mbps 353 Mbps 1.93 5.9M / 4.2M
So while the compression ratio is lower, the bandwidth savings are
much more important due to the significantly lower compression cost
which allows to consume even more data from the servers. In the
example above, zlib became the bottleneck at 24% of the output
bandwidth. Also the difference in memory usage is obvious.
More tests run on a single core of a core i5-3320M, with 500 concurrent
users and the default 16kB buffers :
At 100% CPU (no limit) :
BW In BW Out BW Saved Ratio memory VSZ/RSS hits/s
zlib 480 Mbps 188 Mbps 292 Mbps 2.55 130M / 101M 744
slz 1700 Mbps 810 Mbps 890 Mbps 2.10 23.7M / 9.7M 2382
At 85% CPU (limited) :
BW In BW Out BW Saved Ratio memory VSZ/RSS hits/s
zlib 1240 Mbps 976 Mbps 264 Mbps 1.27 130M / 100M 1738
slz 1600 Mbps 976 Mbps 624 Mbps 1.64 23.7M / 9.7M 2210
The most important benefit really happens when the CPU usage is
limited by "maxcompcpuusage" or the BW limited by "maxcomprate" :
in order to preserve resources, haproxy throttles the compression
ratio until usage is within limits. Since slz is much cheaper, the
average compression ratio is much higher and the input bandwidth
is quite higher for one Gbps output.
Other tests made with some reference files :
BW In BW Out BW Saved Ratio hits/s
daniels.html zlib 1320 Mbps 163 Mbps 1157 Mbps 8.10 1925
slz 3600 Mbps 580 Mbps 3020 Mbps 6.20 5300
tv.com/listing zlib 980 Mbps 124 Mbps 856 Mbps 7.90 310
slz 3300 Mbps 553 Mbps 2747 Mbps 5.97 1100
jquery.min.js zlib 430 Mbps 180 Mbps 250 Mbps 2.39 547
slz 1470 Mbps 764 Mbps 706 Mbps 1.92 1815
bootstrap.min.css zlib 790 Mbps 165 Mbps 625 Mbps 4.79 777
slz 2450 Mbps 650 Mbps 1800 Mbps 3.77 2400
So on top of saving a lot of memory, slz is constantly 2.5-3.5 times
faster than zlib and results in providing more savings for a fixed CPU
usage. For links smaller than 100 Mbps, zlib still provides a better
compression ratio, at the expense of a much higher CPU usage.
Larger input files provide slightly higher bandwidth for both libs, at
the expense of a bit more memory usage for zlib (it converges to 256kB
per connection).
2015-03-28 21:32:06 -04:00
|
|
|
/*
|
|
|
|
|
* Deinit the algorithm
|
|
|
|
|
*/
|
|
|
|
|
static int identity_end(struct comp_ctx **comp_ctx)
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
{
|
|
|
|
|
return 0;
|
|
|
|
|
}
|
|
|
|
|
|
MAJOR: compression: integrate support for libslz
This library is designed to emit a zlib-compatible stream with no
memory usage and to favor resource savings over compression ratio.
While zlib requires 256 kB of RAM per compression context (and can only
support 4000 connections per GB of RAM), the stateless compression
offered by libslz does not need to retain buffers between subsequent
calls. In theory this slightly reduces the compression ratio but in
practice it does not have that much of an effect since the zlib
window is limited to 32kB.
Libslz is available at :
http://git.1wt.eu/web?p=libslz.git
It was designed for web compression and provides a lot of savings
over zlib in haproxy. Here are the preliminary results on a single
core of a core2-quad 3.0 GHz in 32-bit for only 300 concurrent
sessions visiting the home page of www.haproxy.org (76 kB) with
the default 16kB buffers :
BW In BW Out BW Saved Ratio memory VSZ/RSS
zlib 237 Mbps 92 Mbps 145 Mbps 2.58 84M / 69M
slz 733 Mbps 380 Mbps 353 Mbps 1.93 5.9M / 4.2M
So while the compression ratio is lower, the bandwidth savings are
much more important due to the significantly lower compression cost
which allows to consume even more data from the servers. In the
example above, zlib became the bottleneck at 24% of the output
bandwidth. Also the difference in memory usage is obvious.
More tests run on a single core of a core i5-3320M, with 500 concurrent
users and the default 16kB buffers :
At 100% CPU (no limit) :
BW In BW Out BW Saved Ratio memory VSZ/RSS hits/s
zlib 480 Mbps 188 Mbps 292 Mbps 2.55 130M / 101M 744
slz 1700 Mbps 810 Mbps 890 Mbps 2.10 23.7M / 9.7M 2382
At 85% CPU (limited) :
BW In BW Out BW Saved Ratio memory VSZ/RSS hits/s
zlib 1240 Mbps 976 Mbps 264 Mbps 1.27 130M / 100M 1738
slz 1600 Mbps 976 Mbps 624 Mbps 1.64 23.7M / 9.7M 2210
The most important benefit really happens when the CPU usage is
limited by "maxcompcpuusage" or the BW limited by "maxcomprate" :
in order to preserve resources, haproxy throttles the compression
ratio until usage is within limits. Since slz is much cheaper, the
average compression ratio is much higher and the input bandwidth
is quite higher for one Gbps output.
Other tests made with some reference files :
BW In BW Out BW Saved Ratio hits/s
daniels.html zlib 1320 Mbps 163 Mbps 1157 Mbps 8.10 1925
slz 3600 Mbps 580 Mbps 3020 Mbps 6.20 5300
tv.com/listing zlib 980 Mbps 124 Mbps 856 Mbps 7.90 310
slz 3300 Mbps 553 Mbps 2747 Mbps 5.97 1100
jquery.min.js zlib 430 Mbps 180 Mbps 250 Mbps 2.39 547
slz 1470 Mbps 764 Mbps 706 Mbps 1.92 1815
bootstrap.min.css zlib 790 Mbps 165 Mbps 625 Mbps 4.79 777
slz 2450 Mbps 650 Mbps 1800 Mbps 3.77 2400
So on top of saving a lot of memory, slz is constantly 2.5-3.5 times
faster than zlib and results in providing more savings for a fixed CPU
usage. For links smaller than 100 Mbps, zlib still provides a better
compression ratio, at the expense of a much higher CPU usage.
Larger input files provide slightly higher bandwidth for both libs, at
the expense of a bit more memory usage for zlib (it converges to 256kB
per connection).
2015-03-28 21:32:06 -04:00
|
|
|
|
|
|
|
|
#ifdef USE_SLZ
|
|
|
|
|
|
|
|
|
|
/* SLZ's gzip format (RFC1952). Returns < 0 on error. */
|
|
|
|
|
static int rfc1952_init(struct comp_ctx **comp_ctx, int level)
|
|
|
|
|
{
|
|
|
|
|
if (init_comp_ctx(comp_ctx) < 0)
|
|
|
|
|
return -1;
|
|
|
|
|
|
|
|
|
|
(*comp_ctx)->cur_lvl = !!level;
|
|
|
|
|
return slz_rfc1952_init(&(*comp_ctx)->strm, !!level);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* SLZ's raw deflate format (RFC1951). Returns < 0 on error. */
|
|
|
|
|
static int rfc1951_init(struct comp_ctx **comp_ctx, int level)
|
|
|
|
|
{
|
|
|
|
|
if (init_comp_ctx(comp_ctx) < 0)
|
|
|
|
|
return -1;
|
|
|
|
|
|
|
|
|
|
(*comp_ctx)->cur_lvl = !!level;
|
|
|
|
|
return slz_rfc1951_init(&(*comp_ctx)->strm, !!level);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* SLZ's zlib format (RFC1950). Returns < 0 on error. */
|
|
|
|
|
static int rfc1950_init(struct comp_ctx **comp_ctx, int level)
|
|
|
|
|
{
|
|
|
|
|
if (init_comp_ctx(comp_ctx) < 0)
|
|
|
|
|
return -1;
|
|
|
|
|
|
|
|
|
|
(*comp_ctx)->cur_lvl = !!level;
|
|
|
|
|
return slz_rfc1950_init(&(*comp_ctx)->strm, !!level);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* Return the size of consumed data or -1. The output buffer is unused at this
|
|
|
|
|
* point, we only keep a reference to the input data or a copy of them if the
|
|
|
|
|
* reference is already used.
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
*/
|
MAJOR: compression: integrate support for libslz
This library is designed to emit a zlib-compatible stream with no
memory usage and to favor resource savings over compression ratio.
While zlib requires 256 kB of RAM per compression context (and can only
support 4000 connections per GB of RAM), the stateless compression
offered by libslz does not need to retain buffers between subsequent
calls. In theory this slightly reduces the compression ratio but in
practice it does not have that much of an effect since the zlib
window is limited to 32kB.
Libslz is available at :
http://git.1wt.eu/web?p=libslz.git
It was designed for web compression and provides a lot of savings
over zlib in haproxy. Here are the preliminary results on a single
core of a core2-quad 3.0 GHz in 32-bit for only 300 concurrent
sessions visiting the home page of www.haproxy.org (76 kB) with
the default 16kB buffers :
BW In BW Out BW Saved Ratio memory VSZ/RSS
zlib 237 Mbps 92 Mbps 145 Mbps 2.58 84M / 69M
slz 733 Mbps 380 Mbps 353 Mbps 1.93 5.9M / 4.2M
So while the compression ratio is lower, the bandwidth savings are
much more important due to the significantly lower compression cost
which allows to consume even more data from the servers. In the
example above, zlib became the bottleneck at 24% of the output
bandwidth. Also the difference in memory usage is obvious.
More tests run on a single core of a core i5-3320M, with 500 concurrent
users and the default 16kB buffers :
At 100% CPU (no limit) :
BW In BW Out BW Saved Ratio memory VSZ/RSS hits/s
zlib 480 Mbps 188 Mbps 292 Mbps 2.55 130M / 101M 744
slz 1700 Mbps 810 Mbps 890 Mbps 2.10 23.7M / 9.7M 2382
At 85% CPU (limited) :
BW In BW Out BW Saved Ratio memory VSZ/RSS hits/s
zlib 1240 Mbps 976 Mbps 264 Mbps 1.27 130M / 100M 1738
slz 1600 Mbps 976 Mbps 624 Mbps 1.64 23.7M / 9.7M 2210
The most important benefit really happens when the CPU usage is
limited by "maxcompcpuusage" or the BW limited by "maxcomprate" :
in order to preserve resources, haproxy throttles the compression
ratio until usage is within limits. Since slz is much cheaper, the
average compression ratio is much higher and the input bandwidth
is quite higher for one Gbps output.
Other tests made with some reference files :
BW In BW Out BW Saved Ratio hits/s
daniels.html zlib 1320 Mbps 163 Mbps 1157 Mbps 8.10 1925
slz 3600 Mbps 580 Mbps 3020 Mbps 6.20 5300
tv.com/listing zlib 980 Mbps 124 Mbps 856 Mbps 7.90 310
slz 3300 Mbps 553 Mbps 2747 Mbps 5.97 1100
jquery.min.js zlib 430 Mbps 180 Mbps 250 Mbps 2.39 547
slz 1470 Mbps 764 Mbps 706 Mbps 1.92 1815
bootstrap.min.css zlib 790 Mbps 165 Mbps 625 Mbps 4.79 777
slz 2450 Mbps 650 Mbps 1800 Mbps 3.77 2400
So on top of saving a lot of memory, slz is constantly 2.5-3.5 times
faster than zlib and results in providing more savings for a fixed CPU
usage. For links smaller than 100 Mbps, zlib still provides a better
compression ratio, at the expense of a much higher CPU usage.
Larger input files provide slightly higher bandwidth for both libs, at
the expense of a bit more memory usage for zlib (it converges to 256kB
per connection).
2015-03-28 21:32:06 -04:00
|
|
|
static int rfc195x_add_data(struct comp_ctx *comp_ctx, const char *in_data, int in_len, struct buffer *out)
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
{
|
2018-07-10 11:43:27 -04:00
|
|
|
static THREAD_LOCAL struct buffer tmpbuf = BUF_NULL;
|
MAJOR: compression: integrate support for libslz
This library is designed to emit a zlib-compatible stream with no
memory usage and to favor resource savings over compression ratio.
While zlib requires 256 kB of RAM per compression context (and can only
support 4000 connections per GB of RAM), the stateless compression
offered by libslz does not need to retain buffers between subsequent
calls. In theory this slightly reduces the compression ratio but in
practice it does not have that much of an effect since the zlib
window is limited to 32kB.
Libslz is available at :
http://git.1wt.eu/web?p=libslz.git
It was designed for web compression and provides a lot of savings
over zlib in haproxy. Here are the preliminary results on a single
core of a core2-quad 3.0 GHz in 32-bit for only 300 concurrent
sessions visiting the home page of www.haproxy.org (76 kB) with
the default 16kB buffers :
BW In BW Out BW Saved Ratio memory VSZ/RSS
zlib 237 Mbps 92 Mbps 145 Mbps 2.58 84M / 69M
slz 733 Mbps 380 Mbps 353 Mbps 1.93 5.9M / 4.2M
So while the compression ratio is lower, the bandwidth savings are
much more important due to the significantly lower compression cost
which allows to consume even more data from the servers. In the
example above, zlib became the bottleneck at 24% of the output
bandwidth. Also the difference in memory usage is obvious.
More tests run on a single core of a core i5-3320M, with 500 concurrent
users and the default 16kB buffers :
At 100% CPU (no limit) :
BW In BW Out BW Saved Ratio memory VSZ/RSS hits/s
zlib 480 Mbps 188 Mbps 292 Mbps 2.55 130M / 101M 744
slz 1700 Mbps 810 Mbps 890 Mbps 2.10 23.7M / 9.7M 2382
At 85% CPU (limited) :
BW In BW Out BW Saved Ratio memory VSZ/RSS hits/s
zlib 1240 Mbps 976 Mbps 264 Mbps 1.27 130M / 100M 1738
slz 1600 Mbps 976 Mbps 624 Mbps 1.64 23.7M / 9.7M 2210
The most important benefit really happens when the CPU usage is
limited by "maxcompcpuusage" or the BW limited by "maxcomprate" :
in order to preserve resources, haproxy throttles the compression
ratio until usage is within limits. Since slz is much cheaper, the
average compression ratio is much higher and the input bandwidth
is quite higher for one Gbps output.
Other tests made with some reference files :
BW In BW Out BW Saved Ratio hits/s
daniels.html zlib 1320 Mbps 163 Mbps 1157 Mbps 8.10 1925
slz 3600 Mbps 580 Mbps 3020 Mbps 6.20 5300
tv.com/listing zlib 980 Mbps 124 Mbps 856 Mbps 7.90 310
slz 3300 Mbps 553 Mbps 2747 Mbps 5.97 1100
jquery.min.js zlib 430 Mbps 180 Mbps 250 Mbps 2.39 547
slz 1470 Mbps 764 Mbps 706 Mbps 1.92 1815
bootstrap.min.css zlib 790 Mbps 165 Mbps 625 Mbps 4.79 777
slz 2450 Mbps 650 Mbps 1800 Mbps 3.77 2400
So on top of saving a lot of memory, slz is constantly 2.5-3.5 times
faster than zlib and results in providing more savings for a fixed CPU
usage. For links smaller than 100 Mbps, zlib still provides a better
compression ratio, at the expense of a much higher CPU usage.
Larger input files provide slightly higher bandwidth for both libs, at
the expense of a bit more memory usage for zlib (it converges to 256kB
per connection).
2015-03-28 21:32:06 -04:00
|
|
|
|
|
|
|
|
if (in_len <= 0)
|
|
|
|
|
return 0;
|
|
|
|
|
|
2018-07-10 11:43:27 -04:00
|
|
|
if (comp_ctx->direct_ptr && b_is_null(&comp_ctx->queued)) {
|
MAJOR: compression: integrate support for libslz
This library is designed to emit a zlib-compatible stream with no
memory usage and to favor resource savings over compression ratio.
While zlib requires 256 kB of RAM per compression context (and can only
support 4000 connections per GB of RAM), the stateless compression
offered by libslz does not need to retain buffers between subsequent
calls. In theory this slightly reduces the compression ratio but in
practice it does not have that much of an effect since the zlib
window is limited to 32kB.
Libslz is available at :
http://git.1wt.eu/web?p=libslz.git
It was designed for web compression and provides a lot of savings
over zlib in haproxy. Here are the preliminary results on a single
core of a core2-quad 3.0 GHz in 32-bit for only 300 concurrent
sessions visiting the home page of www.haproxy.org (76 kB) with
the default 16kB buffers :
BW In BW Out BW Saved Ratio memory VSZ/RSS
zlib 237 Mbps 92 Mbps 145 Mbps 2.58 84M / 69M
slz 733 Mbps 380 Mbps 353 Mbps 1.93 5.9M / 4.2M
So while the compression ratio is lower, the bandwidth savings are
much more important due to the significantly lower compression cost
which allows to consume even more data from the servers. In the
example above, zlib became the bottleneck at 24% of the output
bandwidth. Also the difference in memory usage is obvious.
More tests run on a single core of a core i5-3320M, with 500 concurrent
users and the default 16kB buffers :
At 100% CPU (no limit) :
BW In BW Out BW Saved Ratio memory VSZ/RSS hits/s
zlib 480 Mbps 188 Mbps 292 Mbps 2.55 130M / 101M 744
slz 1700 Mbps 810 Mbps 890 Mbps 2.10 23.7M / 9.7M 2382
At 85% CPU (limited) :
BW In BW Out BW Saved Ratio memory VSZ/RSS hits/s
zlib 1240 Mbps 976 Mbps 264 Mbps 1.27 130M / 100M 1738
slz 1600 Mbps 976 Mbps 624 Mbps 1.64 23.7M / 9.7M 2210
The most important benefit really happens when the CPU usage is
limited by "maxcompcpuusage" or the BW limited by "maxcomprate" :
in order to preserve resources, haproxy throttles the compression
ratio until usage is within limits. Since slz is much cheaper, the
average compression ratio is much higher and the input bandwidth
is quite higher for one Gbps output.
Other tests made with some reference files :
BW In BW Out BW Saved Ratio hits/s
daniels.html zlib 1320 Mbps 163 Mbps 1157 Mbps 8.10 1925
slz 3600 Mbps 580 Mbps 3020 Mbps 6.20 5300
tv.com/listing zlib 980 Mbps 124 Mbps 856 Mbps 7.90 310
slz 3300 Mbps 553 Mbps 2747 Mbps 5.97 1100
jquery.min.js zlib 430 Mbps 180 Mbps 250 Mbps 2.39 547
slz 1470 Mbps 764 Mbps 706 Mbps 1.92 1815
bootstrap.min.css zlib 790 Mbps 165 Mbps 625 Mbps 4.79 777
slz 2450 Mbps 650 Mbps 1800 Mbps 3.77 2400
So on top of saving a lot of memory, slz is constantly 2.5-3.5 times
faster than zlib and results in providing more savings for a fixed CPU
usage. For links smaller than 100 Mbps, zlib still provides a better
compression ratio, at the expense of a much higher CPU usage.
Larger input files provide slightly higher bandwidth for both libs, at
the expense of a bit more memory usage for zlib (it converges to 256kB
per connection).
2015-03-28 21:32:06 -04:00
|
|
|
/* data already being pointed to, we're in front of fragmented
|
|
|
|
|
* data and need a buffer now. We reuse the same buffer, as it's
|
|
|
|
|
* not used out of the scope of a series of add_data()*, end().
|
|
|
|
|
*/
|
2018-07-10 11:43:27 -04:00
|
|
|
if (unlikely(!tmpbuf.size)) {
|
MAJOR: compression: integrate support for libslz
This library is designed to emit a zlib-compatible stream with no
memory usage and to favor resource savings over compression ratio.
While zlib requires 256 kB of RAM per compression context (and can only
support 4000 connections per GB of RAM), the stateless compression
offered by libslz does not need to retain buffers between subsequent
calls. In theory this slightly reduces the compression ratio but in
practice it does not have that much of an effect since the zlib
window is limited to 32kB.
Libslz is available at :
http://git.1wt.eu/web?p=libslz.git
It was designed for web compression and provides a lot of savings
over zlib in haproxy. Here are the preliminary results on a single
core of a core2-quad 3.0 GHz in 32-bit for only 300 concurrent
sessions visiting the home page of www.haproxy.org (76 kB) with
the default 16kB buffers :
BW In BW Out BW Saved Ratio memory VSZ/RSS
zlib 237 Mbps 92 Mbps 145 Mbps 2.58 84M / 69M
slz 733 Mbps 380 Mbps 353 Mbps 1.93 5.9M / 4.2M
So while the compression ratio is lower, the bandwidth savings are
much more important due to the significantly lower compression cost
which allows to consume even more data from the servers. In the
example above, zlib became the bottleneck at 24% of the output
bandwidth. Also the difference in memory usage is obvious.
More tests run on a single core of a core i5-3320M, with 500 concurrent
users and the default 16kB buffers :
At 100% CPU (no limit) :
BW In BW Out BW Saved Ratio memory VSZ/RSS hits/s
zlib 480 Mbps 188 Mbps 292 Mbps 2.55 130M / 101M 744
slz 1700 Mbps 810 Mbps 890 Mbps 2.10 23.7M / 9.7M 2382
At 85% CPU (limited) :
BW In BW Out BW Saved Ratio memory VSZ/RSS hits/s
zlib 1240 Mbps 976 Mbps 264 Mbps 1.27 130M / 100M 1738
slz 1600 Mbps 976 Mbps 624 Mbps 1.64 23.7M / 9.7M 2210
The most important benefit really happens when the CPU usage is
limited by "maxcompcpuusage" or the BW limited by "maxcomprate" :
in order to preserve resources, haproxy throttles the compression
ratio until usage is within limits. Since slz is much cheaper, the
average compression ratio is much higher and the input bandwidth
is quite higher for one Gbps output.
Other tests made with some reference files :
BW In BW Out BW Saved Ratio hits/s
daniels.html zlib 1320 Mbps 163 Mbps 1157 Mbps 8.10 1925
slz 3600 Mbps 580 Mbps 3020 Mbps 6.20 5300
tv.com/listing zlib 980 Mbps 124 Mbps 856 Mbps 7.90 310
slz 3300 Mbps 553 Mbps 2747 Mbps 5.97 1100
jquery.min.js zlib 430 Mbps 180 Mbps 250 Mbps 2.39 547
slz 1470 Mbps 764 Mbps 706 Mbps 1.92 1815
bootstrap.min.css zlib 790 Mbps 165 Mbps 625 Mbps 4.79 777
slz 2450 Mbps 650 Mbps 1800 Mbps 3.77 2400
So on top of saving a lot of memory, slz is constantly 2.5-3.5 times
faster than zlib and results in providing more savings for a fixed CPU
usage. For links smaller than 100 Mbps, zlib still provides a better
compression ratio, at the expense of a much higher CPU usage.
Larger input files provide slightly higher bandwidth for both libs, at
the expense of a bit more memory usage for zlib (it converges to 256kB
per connection).
2015-03-28 21:32:06 -04:00
|
|
|
/* this is the first time we need the compression buffer */
|
|
|
|
|
if (b_alloc(&tmpbuf) == NULL)
|
|
|
|
|
return -1; /* no memory */
|
|
|
|
|
}
|
2018-07-10 11:43:27 -04:00
|
|
|
b_reset(&tmpbuf);
|
|
|
|
|
memcpy(b_tail(&tmpbuf), comp_ctx->direct_ptr, comp_ctx->direct_len);
|
|
|
|
|
b_add(&tmpbuf, comp_ctx->direct_len);
|
MAJOR: compression: integrate support for libslz
This library is designed to emit a zlib-compatible stream with no
memory usage and to favor resource savings over compression ratio.
While zlib requires 256 kB of RAM per compression context (and can only
support 4000 connections per GB of RAM), the stateless compression
offered by libslz does not need to retain buffers between subsequent
calls. In theory this slightly reduces the compression ratio but in
practice it does not have that much of an effect since the zlib
window is limited to 32kB.
Libslz is available at :
http://git.1wt.eu/web?p=libslz.git
It was designed for web compression and provides a lot of savings
over zlib in haproxy. Here are the preliminary results on a single
core of a core2-quad 3.0 GHz in 32-bit for only 300 concurrent
sessions visiting the home page of www.haproxy.org (76 kB) with
the default 16kB buffers :
BW In BW Out BW Saved Ratio memory VSZ/RSS
zlib 237 Mbps 92 Mbps 145 Mbps 2.58 84M / 69M
slz 733 Mbps 380 Mbps 353 Mbps 1.93 5.9M / 4.2M
So while the compression ratio is lower, the bandwidth savings are
much more important due to the significantly lower compression cost
which allows to consume even more data from the servers. In the
example above, zlib became the bottleneck at 24% of the output
bandwidth. Also the difference in memory usage is obvious.
More tests run on a single core of a core i5-3320M, with 500 concurrent
users and the default 16kB buffers :
At 100% CPU (no limit) :
BW In BW Out BW Saved Ratio memory VSZ/RSS hits/s
zlib 480 Mbps 188 Mbps 292 Mbps 2.55 130M / 101M 744
slz 1700 Mbps 810 Mbps 890 Mbps 2.10 23.7M / 9.7M 2382
At 85% CPU (limited) :
BW In BW Out BW Saved Ratio memory VSZ/RSS hits/s
zlib 1240 Mbps 976 Mbps 264 Mbps 1.27 130M / 100M 1738
slz 1600 Mbps 976 Mbps 624 Mbps 1.64 23.7M / 9.7M 2210
The most important benefit really happens when the CPU usage is
limited by "maxcompcpuusage" or the BW limited by "maxcomprate" :
in order to preserve resources, haproxy throttles the compression
ratio until usage is within limits. Since slz is much cheaper, the
average compression ratio is much higher and the input bandwidth
is quite higher for one Gbps output.
Other tests made with some reference files :
BW In BW Out BW Saved Ratio hits/s
daniels.html zlib 1320 Mbps 163 Mbps 1157 Mbps 8.10 1925
slz 3600 Mbps 580 Mbps 3020 Mbps 6.20 5300
tv.com/listing zlib 980 Mbps 124 Mbps 856 Mbps 7.90 310
slz 3300 Mbps 553 Mbps 2747 Mbps 5.97 1100
jquery.min.js zlib 430 Mbps 180 Mbps 250 Mbps 2.39 547
slz 1470 Mbps 764 Mbps 706 Mbps 1.92 1815
bootstrap.min.css zlib 790 Mbps 165 Mbps 625 Mbps 4.79 777
slz 2450 Mbps 650 Mbps 1800 Mbps 3.77 2400
So on top of saving a lot of memory, slz is constantly 2.5-3.5 times
faster than zlib and results in providing more savings for a fixed CPU
usage. For links smaller than 100 Mbps, zlib still provides a better
compression ratio, at the expense of a much higher CPU usage.
Larger input files provide slightly higher bandwidth for both libs, at
the expense of a bit more memory usage for zlib (it converges to 256kB
per connection).
2015-03-28 21:32:06 -04:00
|
|
|
comp_ctx->direct_ptr = NULL;
|
|
|
|
|
comp_ctx->direct_len = 0;
|
|
|
|
|
comp_ctx->queued = tmpbuf;
|
|
|
|
|
/* fall through buffer copy */
|
|
|
|
|
}
|
|
|
|
|
|
2018-07-10 11:43:27 -04:00
|
|
|
if (!b_is_null(&comp_ctx->queued)) {
|
MAJOR: compression: integrate support for libslz
This library is designed to emit a zlib-compatible stream with no
memory usage and to favor resource savings over compression ratio.
While zlib requires 256 kB of RAM per compression context (and can only
support 4000 connections per GB of RAM), the stateless compression
offered by libslz does not need to retain buffers between subsequent
calls. In theory this slightly reduces the compression ratio but in
practice it does not have that much of an effect since the zlib
window is limited to 32kB.
Libslz is available at :
http://git.1wt.eu/web?p=libslz.git
It was designed for web compression and provides a lot of savings
over zlib in haproxy. Here are the preliminary results on a single
core of a core2-quad 3.0 GHz in 32-bit for only 300 concurrent
sessions visiting the home page of www.haproxy.org (76 kB) with
the default 16kB buffers :
BW In BW Out BW Saved Ratio memory VSZ/RSS
zlib 237 Mbps 92 Mbps 145 Mbps 2.58 84M / 69M
slz 733 Mbps 380 Mbps 353 Mbps 1.93 5.9M / 4.2M
So while the compression ratio is lower, the bandwidth savings are
much more important due to the significantly lower compression cost
which allows to consume even more data from the servers. In the
example above, zlib became the bottleneck at 24% of the output
bandwidth. Also the difference in memory usage is obvious.
More tests run on a single core of a core i5-3320M, with 500 concurrent
users and the default 16kB buffers :
At 100% CPU (no limit) :
BW In BW Out BW Saved Ratio memory VSZ/RSS hits/s
zlib 480 Mbps 188 Mbps 292 Mbps 2.55 130M / 101M 744
slz 1700 Mbps 810 Mbps 890 Mbps 2.10 23.7M / 9.7M 2382
At 85% CPU (limited) :
BW In BW Out BW Saved Ratio memory VSZ/RSS hits/s
zlib 1240 Mbps 976 Mbps 264 Mbps 1.27 130M / 100M 1738
slz 1600 Mbps 976 Mbps 624 Mbps 1.64 23.7M / 9.7M 2210
The most important benefit really happens when the CPU usage is
limited by "maxcompcpuusage" or the BW limited by "maxcomprate" :
in order to preserve resources, haproxy throttles the compression
ratio until usage is within limits. Since slz is much cheaper, the
average compression ratio is much higher and the input bandwidth
is quite higher for one Gbps output.
Other tests made with some reference files :
BW In BW Out BW Saved Ratio hits/s
daniels.html zlib 1320 Mbps 163 Mbps 1157 Mbps 8.10 1925
slz 3600 Mbps 580 Mbps 3020 Mbps 6.20 5300
tv.com/listing zlib 980 Mbps 124 Mbps 856 Mbps 7.90 310
slz 3300 Mbps 553 Mbps 2747 Mbps 5.97 1100
jquery.min.js zlib 430 Mbps 180 Mbps 250 Mbps 2.39 547
slz 1470 Mbps 764 Mbps 706 Mbps 1.92 1815
bootstrap.min.css zlib 790 Mbps 165 Mbps 625 Mbps 4.79 777
slz 2450 Mbps 650 Mbps 1800 Mbps 3.77 2400
So on top of saving a lot of memory, slz is constantly 2.5-3.5 times
faster than zlib and results in providing more savings for a fixed CPU
usage. For links smaller than 100 Mbps, zlib still provides a better
compression ratio, at the expense of a much higher CPU usage.
Larger input files provide slightly higher bandwidth for both libs, at
the expense of a bit more memory usage for zlib (it converges to 256kB
per connection).
2015-03-28 21:32:06 -04:00
|
|
|
/* data already pending */
|
2018-07-10 11:43:27 -04:00
|
|
|
memcpy(b_tail(&comp_ctx->queued), in_data, in_len);
|
|
|
|
|
b_add(&comp_ctx->queued, in_len);
|
MAJOR: compression: integrate support for libslz
This library is designed to emit a zlib-compatible stream with no
memory usage and to favor resource savings over compression ratio.
While zlib requires 256 kB of RAM per compression context (and can only
support 4000 connections per GB of RAM), the stateless compression
offered by libslz does not need to retain buffers between subsequent
calls. In theory this slightly reduces the compression ratio but in
practice it does not have that much of an effect since the zlib
window is limited to 32kB.
Libslz is available at :
http://git.1wt.eu/web?p=libslz.git
It was designed for web compression and provides a lot of savings
over zlib in haproxy. Here are the preliminary results on a single
core of a core2-quad 3.0 GHz in 32-bit for only 300 concurrent
sessions visiting the home page of www.haproxy.org (76 kB) with
the default 16kB buffers :
BW In BW Out BW Saved Ratio memory VSZ/RSS
zlib 237 Mbps 92 Mbps 145 Mbps 2.58 84M / 69M
slz 733 Mbps 380 Mbps 353 Mbps 1.93 5.9M / 4.2M
So while the compression ratio is lower, the bandwidth savings are
much more important due to the significantly lower compression cost
which allows to consume even more data from the servers. In the
example above, zlib became the bottleneck at 24% of the output
bandwidth. Also the difference in memory usage is obvious.
More tests run on a single core of a core i5-3320M, with 500 concurrent
users and the default 16kB buffers :
At 100% CPU (no limit) :
BW In BW Out BW Saved Ratio memory VSZ/RSS hits/s
zlib 480 Mbps 188 Mbps 292 Mbps 2.55 130M / 101M 744
slz 1700 Mbps 810 Mbps 890 Mbps 2.10 23.7M / 9.7M 2382
At 85% CPU (limited) :
BW In BW Out BW Saved Ratio memory VSZ/RSS hits/s
zlib 1240 Mbps 976 Mbps 264 Mbps 1.27 130M / 100M 1738
slz 1600 Mbps 976 Mbps 624 Mbps 1.64 23.7M / 9.7M 2210
The most important benefit really happens when the CPU usage is
limited by "maxcompcpuusage" or the BW limited by "maxcomprate" :
in order to preserve resources, haproxy throttles the compression
ratio until usage is within limits. Since slz is much cheaper, the
average compression ratio is much higher and the input bandwidth
is quite higher for one Gbps output.
Other tests made with some reference files :
BW In BW Out BW Saved Ratio hits/s
daniels.html zlib 1320 Mbps 163 Mbps 1157 Mbps 8.10 1925
slz 3600 Mbps 580 Mbps 3020 Mbps 6.20 5300
tv.com/listing zlib 980 Mbps 124 Mbps 856 Mbps 7.90 310
slz 3300 Mbps 553 Mbps 2747 Mbps 5.97 1100
jquery.min.js zlib 430 Mbps 180 Mbps 250 Mbps 2.39 547
slz 1470 Mbps 764 Mbps 706 Mbps 1.92 1815
bootstrap.min.css zlib 790 Mbps 165 Mbps 625 Mbps 4.79 777
slz 2450 Mbps 650 Mbps 1800 Mbps 3.77 2400
So on top of saving a lot of memory, slz is constantly 2.5-3.5 times
faster than zlib and results in providing more savings for a fixed CPU
usage. For links smaller than 100 Mbps, zlib still provides a better
compression ratio, at the expense of a much higher CPU usage.
Larger input files provide slightly higher bandwidth for both libs, at
the expense of a bit more memory usage for zlib (it converges to 256kB
per connection).
2015-03-28 21:32:06 -04:00
|
|
|
return in_len;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
comp_ctx->direct_ptr = in_data;
|
|
|
|
|
comp_ctx->direct_len = in_len;
|
|
|
|
|
return in_len;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* Compresses the data accumulated using add_data(), and optionally sends the
|
|
|
|
|
* format-specific trailer if <finish> is non-null. <out> is expected to have a
|
|
|
|
|
* large enough free non-wrapping space as verified by http_comp_buffer_init().
|
|
|
|
|
* The number of bytes emitted is reported.
|
|
|
|
|
*/
|
|
|
|
|
static int rfc195x_flush_or_finish(struct comp_ctx *comp_ctx, struct buffer *out, int finish)
|
|
|
|
|
{
|
|
|
|
|
struct slz_stream *strm = &comp_ctx->strm;
|
|
|
|
|
const char *in_ptr;
|
|
|
|
|
int in_len;
|
|
|
|
|
int out_len;
|
|
|
|
|
|
|
|
|
|
in_ptr = comp_ctx->direct_ptr;
|
|
|
|
|
in_len = comp_ctx->direct_len;
|
|
|
|
|
|
2018-07-10 11:43:27 -04:00
|
|
|
if (!b_is_null(&comp_ctx->queued)) {
|
|
|
|
|
in_ptr = b_head(&comp_ctx->queued);
|
|
|
|
|
in_len = b_data(&comp_ctx->queued);
|
MAJOR: compression: integrate support for libslz
This library is designed to emit a zlib-compatible stream with no
memory usage and to favor resource savings over compression ratio.
While zlib requires 256 kB of RAM per compression context (and can only
support 4000 connections per GB of RAM), the stateless compression
offered by libslz does not need to retain buffers between subsequent
calls. In theory this slightly reduces the compression ratio but in
practice it does not have that much of an effect since the zlib
window is limited to 32kB.
Libslz is available at :
http://git.1wt.eu/web?p=libslz.git
It was designed for web compression and provides a lot of savings
over zlib in haproxy. Here are the preliminary results on a single
core of a core2-quad 3.0 GHz in 32-bit for only 300 concurrent
sessions visiting the home page of www.haproxy.org (76 kB) with
the default 16kB buffers :
BW In BW Out BW Saved Ratio memory VSZ/RSS
zlib 237 Mbps 92 Mbps 145 Mbps 2.58 84M / 69M
slz 733 Mbps 380 Mbps 353 Mbps 1.93 5.9M / 4.2M
So while the compression ratio is lower, the bandwidth savings are
much more important due to the significantly lower compression cost
which allows to consume even more data from the servers. In the
example above, zlib became the bottleneck at 24% of the output
bandwidth. Also the difference in memory usage is obvious.
More tests run on a single core of a core i5-3320M, with 500 concurrent
users and the default 16kB buffers :
At 100% CPU (no limit) :
BW In BW Out BW Saved Ratio memory VSZ/RSS hits/s
zlib 480 Mbps 188 Mbps 292 Mbps 2.55 130M / 101M 744
slz 1700 Mbps 810 Mbps 890 Mbps 2.10 23.7M / 9.7M 2382
At 85% CPU (limited) :
BW In BW Out BW Saved Ratio memory VSZ/RSS hits/s
zlib 1240 Mbps 976 Mbps 264 Mbps 1.27 130M / 100M 1738
slz 1600 Mbps 976 Mbps 624 Mbps 1.64 23.7M / 9.7M 2210
The most important benefit really happens when the CPU usage is
limited by "maxcompcpuusage" or the BW limited by "maxcomprate" :
in order to preserve resources, haproxy throttles the compression
ratio until usage is within limits. Since slz is much cheaper, the
average compression ratio is much higher and the input bandwidth
is quite higher for one Gbps output.
Other tests made with some reference files :
BW In BW Out BW Saved Ratio hits/s
daniels.html zlib 1320 Mbps 163 Mbps 1157 Mbps 8.10 1925
slz 3600 Mbps 580 Mbps 3020 Mbps 6.20 5300
tv.com/listing zlib 980 Mbps 124 Mbps 856 Mbps 7.90 310
slz 3300 Mbps 553 Mbps 2747 Mbps 5.97 1100
jquery.min.js zlib 430 Mbps 180 Mbps 250 Mbps 2.39 547
slz 1470 Mbps 764 Mbps 706 Mbps 1.92 1815
bootstrap.min.css zlib 790 Mbps 165 Mbps 625 Mbps 4.79 777
slz 2450 Mbps 650 Mbps 1800 Mbps 3.77 2400
So on top of saving a lot of memory, slz is constantly 2.5-3.5 times
faster than zlib and results in providing more savings for a fixed CPU
usage. For links smaller than 100 Mbps, zlib still provides a better
compression ratio, at the expense of a much higher CPU usage.
Larger input files provide slightly higher bandwidth for both libs, at
the expense of a bit more memory usage for zlib (it converges to 256kB
per connection).
2015-03-28 21:32:06 -04:00
|
|
|
}
|
|
|
|
|
|
2018-06-29 12:16:31 -04:00
|
|
|
out_len = b_data(out);
|
MAJOR: compression: integrate support for libslz
This library is designed to emit a zlib-compatible stream with no
memory usage and to favor resource savings over compression ratio.
While zlib requires 256 kB of RAM per compression context (and can only
support 4000 connections per GB of RAM), the stateless compression
offered by libslz does not need to retain buffers between subsequent
calls. In theory this slightly reduces the compression ratio but in
practice it does not have that much of an effect since the zlib
window is limited to 32kB.
Libslz is available at :
http://git.1wt.eu/web?p=libslz.git
It was designed for web compression and provides a lot of savings
over zlib in haproxy. Here are the preliminary results on a single
core of a core2-quad 3.0 GHz in 32-bit for only 300 concurrent
sessions visiting the home page of www.haproxy.org (76 kB) with
the default 16kB buffers :
BW In BW Out BW Saved Ratio memory VSZ/RSS
zlib 237 Mbps 92 Mbps 145 Mbps 2.58 84M / 69M
slz 733 Mbps 380 Mbps 353 Mbps 1.93 5.9M / 4.2M
So while the compression ratio is lower, the bandwidth savings are
much more important due to the significantly lower compression cost
which allows to consume even more data from the servers. In the
example above, zlib became the bottleneck at 24% of the output
bandwidth. Also the difference in memory usage is obvious.
More tests run on a single core of a core i5-3320M, with 500 concurrent
users and the default 16kB buffers :
At 100% CPU (no limit) :
BW In BW Out BW Saved Ratio memory VSZ/RSS hits/s
zlib 480 Mbps 188 Mbps 292 Mbps 2.55 130M / 101M 744
slz 1700 Mbps 810 Mbps 890 Mbps 2.10 23.7M / 9.7M 2382
At 85% CPU (limited) :
BW In BW Out BW Saved Ratio memory VSZ/RSS hits/s
zlib 1240 Mbps 976 Mbps 264 Mbps 1.27 130M / 100M 1738
slz 1600 Mbps 976 Mbps 624 Mbps 1.64 23.7M / 9.7M 2210
The most important benefit really happens when the CPU usage is
limited by "maxcompcpuusage" or the BW limited by "maxcomprate" :
in order to preserve resources, haproxy throttles the compression
ratio until usage is within limits. Since slz is much cheaper, the
average compression ratio is much higher and the input bandwidth
is quite higher for one Gbps output.
Other tests made with some reference files :
BW In BW Out BW Saved Ratio hits/s
daniels.html zlib 1320 Mbps 163 Mbps 1157 Mbps 8.10 1925
slz 3600 Mbps 580 Mbps 3020 Mbps 6.20 5300
tv.com/listing zlib 980 Mbps 124 Mbps 856 Mbps 7.90 310
slz 3300 Mbps 553 Mbps 2747 Mbps 5.97 1100
jquery.min.js zlib 430 Mbps 180 Mbps 250 Mbps 2.39 547
slz 1470 Mbps 764 Mbps 706 Mbps 1.92 1815
bootstrap.min.css zlib 790 Mbps 165 Mbps 625 Mbps 4.79 777
slz 2450 Mbps 650 Mbps 1800 Mbps 3.77 2400
So on top of saving a lot of memory, slz is constantly 2.5-3.5 times
faster than zlib and results in providing more savings for a fixed CPU
usage. For links smaller than 100 Mbps, zlib still provides a better
compression ratio, at the expense of a much higher CPU usage.
Larger input files provide slightly higher bandwidth for both libs, at
the expense of a bit more memory usage for zlib (it converges to 256kB
per connection).
2015-03-28 21:32:06 -04:00
|
|
|
|
|
|
|
|
if (in_ptr)
|
2018-06-28 12:17:23 -04:00
|
|
|
b_add(out, slz_encode(strm, b_tail(out), in_ptr, in_len, !finish));
|
MAJOR: compression: integrate support for libslz
This library is designed to emit a zlib-compatible stream with no
memory usage and to favor resource savings over compression ratio.
While zlib requires 256 kB of RAM per compression context (and can only
support 4000 connections per GB of RAM), the stateless compression
offered by libslz does not need to retain buffers between subsequent
calls. In theory this slightly reduces the compression ratio but in
practice it does not have that much of an effect since the zlib
window is limited to 32kB.
Libslz is available at :
http://git.1wt.eu/web?p=libslz.git
It was designed for web compression and provides a lot of savings
over zlib in haproxy. Here are the preliminary results on a single
core of a core2-quad 3.0 GHz in 32-bit for only 300 concurrent
sessions visiting the home page of www.haproxy.org (76 kB) with
the default 16kB buffers :
BW In BW Out BW Saved Ratio memory VSZ/RSS
zlib 237 Mbps 92 Mbps 145 Mbps 2.58 84M / 69M
slz 733 Mbps 380 Mbps 353 Mbps 1.93 5.9M / 4.2M
So while the compression ratio is lower, the bandwidth savings are
much more important due to the significantly lower compression cost
which allows to consume even more data from the servers. In the
example above, zlib became the bottleneck at 24% of the output
bandwidth. Also the difference in memory usage is obvious.
More tests run on a single core of a core i5-3320M, with 500 concurrent
users and the default 16kB buffers :
At 100% CPU (no limit) :
BW In BW Out BW Saved Ratio memory VSZ/RSS hits/s
zlib 480 Mbps 188 Mbps 292 Mbps 2.55 130M / 101M 744
slz 1700 Mbps 810 Mbps 890 Mbps 2.10 23.7M / 9.7M 2382
At 85% CPU (limited) :
BW In BW Out BW Saved Ratio memory VSZ/RSS hits/s
zlib 1240 Mbps 976 Mbps 264 Mbps 1.27 130M / 100M 1738
slz 1600 Mbps 976 Mbps 624 Mbps 1.64 23.7M / 9.7M 2210
The most important benefit really happens when the CPU usage is
limited by "maxcompcpuusage" or the BW limited by "maxcomprate" :
in order to preserve resources, haproxy throttles the compression
ratio until usage is within limits. Since slz is much cheaper, the
average compression ratio is much higher and the input bandwidth
is quite higher for one Gbps output.
Other tests made with some reference files :
BW In BW Out BW Saved Ratio hits/s
daniels.html zlib 1320 Mbps 163 Mbps 1157 Mbps 8.10 1925
slz 3600 Mbps 580 Mbps 3020 Mbps 6.20 5300
tv.com/listing zlib 980 Mbps 124 Mbps 856 Mbps 7.90 310
slz 3300 Mbps 553 Mbps 2747 Mbps 5.97 1100
jquery.min.js zlib 430 Mbps 180 Mbps 250 Mbps 2.39 547
slz 1470 Mbps 764 Mbps 706 Mbps 1.92 1815
bootstrap.min.css zlib 790 Mbps 165 Mbps 625 Mbps 4.79 777
slz 2450 Mbps 650 Mbps 1800 Mbps 3.77 2400
So on top of saving a lot of memory, slz is constantly 2.5-3.5 times
faster than zlib and results in providing more savings for a fixed CPU
usage. For links smaller than 100 Mbps, zlib still provides a better
compression ratio, at the expense of a much higher CPU usage.
Larger input files provide slightly higher bandwidth for both libs, at
the expense of a bit more memory usage for zlib (it converges to 256kB
per connection).
2015-03-28 21:32:06 -04:00
|
|
|
|
|
|
|
|
if (finish)
|
2018-06-28 12:17:23 -04:00
|
|
|
b_add(out, slz_finish(strm, b_tail(out)));
|
MAJOR: compression: integrate support for libslz
This library is designed to emit a zlib-compatible stream with no
memory usage and to favor resource savings over compression ratio.
While zlib requires 256 kB of RAM per compression context (and can only
support 4000 connections per GB of RAM), the stateless compression
offered by libslz does not need to retain buffers between subsequent
calls. In theory this slightly reduces the compression ratio but in
practice it does not have that much of an effect since the zlib
window is limited to 32kB.
Libslz is available at :
http://git.1wt.eu/web?p=libslz.git
It was designed for web compression and provides a lot of savings
over zlib in haproxy. Here are the preliminary results on a single
core of a core2-quad 3.0 GHz in 32-bit for only 300 concurrent
sessions visiting the home page of www.haproxy.org (76 kB) with
the default 16kB buffers :
BW In BW Out BW Saved Ratio memory VSZ/RSS
zlib 237 Mbps 92 Mbps 145 Mbps 2.58 84M / 69M
slz 733 Mbps 380 Mbps 353 Mbps 1.93 5.9M / 4.2M
So while the compression ratio is lower, the bandwidth savings are
much more important due to the significantly lower compression cost
which allows to consume even more data from the servers. In the
example above, zlib became the bottleneck at 24% of the output
bandwidth. Also the difference in memory usage is obvious.
More tests run on a single core of a core i5-3320M, with 500 concurrent
users and the default 16kB buffers :
At 100% CPU (no limit) :
BW In BW Out BW Saved Ratio memory VSZ/RSS hits/s
zlib 480 Mbps 188 Mbps 292 Mbps 2.55 130M / 101M 744
slz 1700 Mbps 810 Mbps 890 Mbps 2.10 23.7M / 9.7M 2382
At 85% CPU (limited) :
BW In BW Out BW Saved Ratio memory VSZ/RSS hits/s
zlib 1240 Mbps 976 Mbps 264 Mbps 1.27 130M / 100M 1738
slz 1600 Mbps 976 Mbps 624 Mbps 1.64 23.7M / 9.7M 2210
The most important benefit really happens when the CPU usage is
limited by "maxcompcpuusage" or the BW limited by "maxcomprate" :
in order to preserve resources, haproxy throttles the compression
ratio until usage is within limits. Since slz is much cheaper, the
average compression ratio is much higher and the input bandwidth
is quite higher for one Gbps output.
Other tests made with some reference files :
BW In BW Out BW Saved Ratio hits/s
daniels.html zlib 1320 Mbps 163 Mbps 1157 Mbps 8.10 1925
slz 3600 Mbps 580 Mbps 3020 Mbps 6.20 5300
tv.com/listing zlib 980 Mbps 124 Mbps 856 Mbps 7.90 310
slz 3300 Mbps 553 Mbps 2747 Mbps 5.97 1100
jquery.min.js zlib 430 Mbps 180 Mbps 250 Mbps 2.39 547
slz 1470 Mbps 764 Mbps 706 Mbps 1.92 1815
bootstrap.min.css zlib 790 Mbps 165 Mbps 625 Mbps 4.79 777
slz 2450 Mbps 650 Mbps 1800 Mbps 3.77 2400
So on top of saving a lot of memory, slz is constantly 2.5-3.5 times
faster than zlib and results in providing more savings for a fixed CPU
usage. For links smaller than 100 Mbps, zlib still provides a better
compression ratio, at the expense of a much higher CPU usage.
Larger input files provide slightly higher bandwidth for both libs, at
the expense of a bit more memory usage for zlib (it converges to 256kB
per connection).
2015-03-28 21:32:06 -04:00
|
|
|
|
2018-06-29 12:16:31 -04:00
|
|
|
out_len = b_data(out) - out_len;
|
MAJOR: compression: integrate support for libslz
This library is designed to emit a zlib-compatible stream with no
memory usage and to favor resource savings over compression ratio.
While zlib requires 256 kB of RAM per compression context (and can only
support 4000 connections per GB of RAM), the stateless compression
offered by libslz does not need to retain buffers between subsequent
calls. In theory this slightly reduces the compression ratio but in
practice it does not have that much of an effect since the zlib
window is limited to 32kB.
Libslz is available at :
http://git.1wt.eu/web?p=libslz.git
It was designed for web compression and provides a lot of savings
over zlib in haproxy. Here are the preliminary results on a single
core of a core2-quad 3.0 GHz in 32-bit for only 300 concurrent
sessions visiting the home page of www.haproxy.org (76 kB) with
the default 16kB buffers :
BW In BW Out BW Saved Ratio memory VSZ/RSS
zlib 237 Mbps 92 Mbps 145 Mbps 2.58 84M / 69M
slz 733 Mbps 380 Mbps 353 Mbps 1.93 5.9M / 4.2M
So while the compression ratio is lower, the bandwidth savings are
much more important due to the significantly lower compression cost
which allows to consume even more data from the servers. In the
example above, zlib became the bottleneck at 24% of the output
bandwidth. Also the difference in memory usage is obvious.
More tests run on a single core of a core i5-3320M, with 500 concurrent
users and the default 16kB buffers :
At 100% CPU (no limit) :
BW In BW Out BW Saved Ratio memory VSZ/RSS hits/s
zlib 480 Mbps 188 Mbps 292 Mbps 2.55 130M / 101M 744
slz 1700 Mbps 810 Mbps 890 Mbps 2.10 23.7M / 9.7M 2382
At 85% CPU (limited) :
BW In BW Out BW Saved Ratio memory VSZ/RSS hits/s
zlib 1240 Mbps 976 Mbps 264 Mbps 1.27 130M / 100M 1738
slz 1600 Mbps 976 Mbps 624 Mbps 1.64 23.7M / 9.7M 2210
The most important benefit really happens when the CPU usage is
limited by "maxcompcpuusage" or the BW limited by "maxcomprate" :
in order to preserve resources, haproxy throttles the compression
ratio until usage is within limits. Since slz is much cheaper, the
average compression ratio is much higher and the input bandwidth
is quite higher for one Gbps output.
Other tests made with some reference files :
BW In BW Out BW Saved Ratio hits/s
daniels.html zlib 1320 Mbps 163 Mbps 1157 Mbps 8.10 1925
slz 3600 Mbps 580 Mbps 3020 Mbps 6.20 5300
tv.com/listing zlib 980 Mbps 124 Mbps 856 Mbps 7.90 310
slz 3300 Mbps 553 Mbps 2747 Mbps 5.97 1100
jquery.min.js zlib 430 Mbps 180 Mbps 250 Mbps 2.39 547
slz 1470 Mbps 764 Mbps 706 Mbps 1.92 1815
bootstrap.min.css zlib 790 Mbps 165 Mbps 625 Mbps 4.79 777
slz 2450 Mbps 650 Mbps 1800 Mbps 3.77 2400
So on top of saving a lot of memory, slz is constantly 2.5-3.5 times
faster than zlib and results in providing more savings for a fixed CPU
usage. For links smaller than 100 Mbps, zlib still provides a better
compression ratio, at the expense of a much higher CPU usage.
Larger input files provide slightly higher bandwidth for both libs, at
the expense of a bit more memory usage for zlib (it converges to 256kB
per connection).
2015-03-28 21:32:06 -04:00
|
|
|
|
|
|
|
|
/* very important, we must wipe the data we've just flushed */
|
|
|
|
|
comp_ctx->direct_len = 0;
|
|
|
|
|
comp_ctx->direct_ptr = NULL;
|
2018-07-10 11:43:27 -04:00
|
|
|
comp_ctx->queued = BUF_NULL;
|
MAJOR: compression: integrate support for libslz
This library is designed to emit a zlib-compatible stream with no
memory usage and to favor resource savings over compression ratio.
While zlib requires 256 kB of RAM per compression context (and can only
support 4000 connections per GB of RAM), the stateless compression
offered by libslz does not need to retain buffers between subsequent
calls. In theory this slightly reduces the compression ratio but in
practice it does not have that much of an effect since the zlib
window is limited to 32kB.
Libslz is available at :
http://git.1wt.eu/web?p=libslz.git
It was designed for web compression and provides a lot of savings
over zlib in haproxy. Here are the preliminary results on a single
core of a core2-quad 3.0 GHz in 32-bit for only 300 concurrent
sessions visiting the home page of www.haproxy.org (76 kB) with
the default 16kB buffers :
BW In BW Out BW Saved Ratio memory VSZ/RSS
zlib 237 Mbps 92 Mbps 145 Mbps 2.58 84M / 69M
slz 733 Mbps 380 Mbps 353 Mbps 1.93 5.9M / 4.2M
So while the compression ratio is lower, the bandwidth savings are
much more important due to the significantly lower compression cost
which allows to consume even more data from the servers. In the
example above, zlib became the bottleneck at 24% of the output
bandwidth. Also the difference in memory usage is obvious.
More tests run on a single core of a core i5-3320M, with 500 concurrent
users and the default 16kB buffers :
At 100% CPU (no limit) :
BW In BW Out BW Saved Ratio memory VSZ/RSS hits/s
zlib 480 Mbps 188 Mbps 292 Mbps 2.55 130M / 101M 744
slz 1700 Mbps 810 Mbps 890 Mbps 2.10 23.7M / 9.7M 2382
At 85% CPU (limited) :
BW In BW Out BW Saved Ratio memory VSZ/RSS hits/s
zlib 1240 Mbps 976 Mbps 264 Mbps 1.27 130M / 100M 1738
slz 1600 Mbps 976 Mbps 624 Mbps 1.64 23.7M / 9.7M 2210
The most important benefit really happens when the CPU usage is
limited by "maxcompcpuusage" or the BW limited by "maxcomprate" :
in order to preserve resources, haproxy throttles the compression
ratio until usage is within limits. Since slz is much cheaper, the
average compression ratio is much higher and the input bandwidth
is quite higher for one Gbps output.
Other tests made with some reference files :
BW In BW Out BW Saved Ratio hits/s
daniels.html zlib 1320 Mbps 163 Mbps 1157 Mbps 8.10 1925
slz 3600 Mbps 580 Mbps 3020 Mbps 6.20 5300
tv.com/listing zlib 980 Mbps 124 Mbps 856 Mbps 7.90 310
slz 3300 Mbps 553 Mbps 2747 Mbps 5.97 1100
jquery.min.js zlib 430 Mbps 180 Mbps 250 Mbps 2.39 547
slz 1470 Mbps 764 Mbps 706 Mbps 1.92 1815
bootstrap.min.css zlib 790 Mbps 165 Mbps 625 Mbps 4.79 777
slz 2450 Mbps 650 Mbps 1800 Mbps 3.77 2400
So on top of saving a lot of memory, slz is constantly 2.5-3.5 times
faster than zlib and results in providing more savings for a fixed CPU
usage. For links smaller than 100 Mbps, zlib still provides a better
compression ratio, at the expense of a much higher CPU usage.
Larger input files provide slightly higher bandwidth for both libs, at
the expense of a bit more memory usage for zlib (it converges to 256kB
per connection).
2015-03-28 21:32:06 -04:00
|
|
|
|
|
|
|
|
/* Verify compression rate limiting and CPU usage */
|
|
|
|
|
if ((global.comp_rate_lim > 0 && (read_freq_ctr(&global.comp_bps_out) > global.comp_rate_lim)) || /* rate */
|
|
|
|
|
(idle_pct < compress_min_idle)) { /* idle */
|
|
|
|
|
if (comp_ctx->cur_lvl > 0)
|
|
|
|
|
strm->level = --comp_ctx->cur_lvl;
|
|
|
|
|
}
|
|
|
|
|
else if (comp_ctx->cur_lvl < global.tune.comp_maxlevel && comp_ctx->cur_lvl < 1) {
|
|
|
|
|
strm->level = ++comp_ctx->cur_lvl;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* and that's all */
|
|
|
|
|
return out_len;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static int rfc195x_flush(struct comp_ctx *comp_ctx, struct buffer *out)
|
|
|
|
|
{
|
|
|
|
|
return rfc195x_flush_or_finish(comp_ctx, out, 0);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static int rfc195x_finish(struct comp_ctx *comp_ctx, struct buffer *out)
|
|
|
|
|
{
|
|
|
|
|
return rfc195x_flush_or_finish(comp_ctx, out, 1);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* we just need to free the comp_ctx here, nothing was allocated */
|
|
|
|
|
static int rfc195x_end(struct comp_ctx **comp_ctx)
|
|
|
|
|
{
|
|
|
|
|
deinit_comp_ctx(comp_ctx);
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
return 0;
|
|
|
|
|
}
|
|
|
|
|
|
MAJOR: compression: integrate support for libslz
This library is designed to emit a zlib-compatible stream with no
memory usage and to favor resource savings over compression ratio.
While zlib requires 256 kB of RAM per compression context (and can only
support 4000 connections per GB of RAM), the stateless compression
offered by libslz does not need to retain buffers between subsequent
calls. In theory this slightly reduces the compression ratio but in
practice it does not have that much of an effect since the zlib
window is limited to 32kB.
Libslz is available at :
http://git.1wt.eu/web?p=libslz.git
It was designed for web compression and provides a lot of savings
over zlib in haproxy. Here are the preliminary results on a single
core of a core2-quad 3.0 GHz in 32-bit for only 300 concurrent
sessions visiting the home page of www.haproxy.org (76 kB) with
the default 16kB buffers :
BW In BW Out BW Saved Ratio memory VSZ/RSS
zlib 237 Mbps 92 Mbps 145 Mbps 2.58 84M / 69M
slz 733 Mbps 380 Mbps 353 Mbps 1.93 5.9M / 4.2M
So while the compression ratio is lower, the bandwidth savings are
much more important due to the significantly lower compression cost
which allows to consume even more data from the servers. In the
example above, zlib became the bottleneck at 24% of the output
bandwidth. Also the difference in memory usage is obvious.
More tests run on a single core of a core i5-3320M, with 500 concurrent
users and the default 16kB buffers :
At 100% CPU (no limit) :
BW In BW Out BW Saved Ratio memory VSZ/RSS hits/s
zlib 480 Mbps 188 Mbps 292 Mbps 2.55 130M / 101M 744
slz 1700 Mbps 810 Mbps 890 Mbps 2.10 23.7M / 9.7M 2382
At 85% CPU (limited) :
BW In BW Out BW Saved Ratio memory VSZ/RSS hits/s
zlib 1240 Mbps 976 Mbps 264 Mbps 1.27 130M / 100M 1738
slz 1600 Mbps 976 Mbps 624 Mbps 1.64 23.7M / 9.7M 2210
The most important benefit really happens when the CPU usage is
limited by "maxcompcpuusage" or the BW limited by "maxcomprate" :
in order to preserve resources, haproxy throttles the compression
ratio until usage is within limits. Since slz is much cheaper, the
average compression ratio is much higher and the input bandwidth
is quite higher for one Gbps output.
Other tests made with some reference files :
BW In BW Out BW Saved Ratio hits/s
daniels.html zlib 1320 Mbps 163 Mbps 1157 Mbps 8.10 1925
slz 3600 Mbps 580 Mbps 3020 Mbps 6.20 5300
tv.com/listing zlib 980 Mbps 124 Mbps 856 Mbps 7.90 310
slz 3300 Mbps 553 Mbps 2747 Mbps 5.97 1100
jquery.min.js zlib 430 Mbps 180 Mbps 250 Mbps 2.39 547
slz 1470 Mbps 764 Mbps 706 Mbps 1.92 1815
bootstrap.min.css zlib 790 Mbps 165 Mbps 625 Mbps 4.79 777
slz 2450 Mbps 650 Mbps 1800 Mbps 3.77 2400
So on top of saving a lot of memory, slz is constantly 2.5-3.5 times
faster than zlib and results in providing more savings for a fixed CPU
usage. For links smaller than 100 Mbps, zlib still provides a better
compression ratio, at the expense of a much higher CPU usage.
Larger input files provide slightly higher bandwidth for both libs, at
the expense of a bit more memory usage for zlib (it converges to 256kB
per connection).
2015-03-28 21:32:06 -04:00
|
|
|
#elif defined(USE_ZLIB) /* ! USE_SLZ */
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
|
2012-10-30 09:30:39 -04:00
|
|
|
/*
|
|
|
|
|
* This is a tricky allocation function using the zlib.
|
|
|
|
|
* This is based on the allocation order in deflateInit2.
|
|
|
|
|
*/
|
|
|
|
|
static void *alloc_zlib(void *opaque, unsigned int items, unsigned int size)
|
|
|
|
|
{
|
|
|
|
|
struct comp_ctx *ctx = opaque;
|
2017-07-25 05:07:15 -04:00
|
|
|
static THREAD_LOCAL char round = 0; /* order in deflateInit2 */
|
2012-10-30 09:30:39 -04:00
|
|
|
void *buf = NULL;
|
2014-12-24 12:07:55 -05:00
|
|
|
struct pool_head *pool = NULL;
|
2012-10-30 09:30:39 -04:00
|
|
|
|
2012-11-20 05:25:20 -05:00
|
|
|
if (global.maxzlibmem > 0 && (global.maxzlibmem - zlib_used_memory) < (long)(items * size))
|
2012-11-07 10:12:57 -05:00
|
|
|
goto end;
|
|
|
|
|
|
2012-10-30 09:30:39 -04:00
|
|
|
switch (round) {
|
|
|
|
|
case 0:
|
2017-07-25 05:07:15 -04:00
|
|
|
if (zlib_pool_deflate_state == NULL) {
|
2017-11-07 04:42:54 -05:00
|
|
|
HA_SPIN_LOCK(COMP_POOL_LOCK, &comp_pool_lock);
|
2017-07-25 05:07:15 -04:00
|
|
|
if (zlib_pool_deflate_state == NULL)
|
|
|
|
|
zlib_pool_deflate_state = create_pool("zlib_state", size * items, MEM_F_SHARED);
|
2017-11-07 04:42:54 -05:00
|
|
|
HA_SPIN_UNLOCK(COMP_POOL_LOCK, &comp_pool_lock);
|
2017-07-25 05:07:15 -04:00
|
|
|
}
|
2014-12-24 12:07:55 -05:00
|
|
|
pool = zlib_pool_deflate_state;
|
2017-11-24 11:34:44 -05:00
|
|
|
ctx->zlib_deflate_state = buf = pool_alloc(pool);
|
2012-10-30 09:30:39 -04:00
|
|
|
break;
|
|
|
|
|
|
|
|
|
|
case 1:
|
2017-07-25 05:07:15 -04:00
|
|
|
if (zlib_pool_window == NULL) {
|
2017-11-07 04:42:54 -05:00
|
|
|
HA_SPIN_LOCK(COMP_POOL_LOCK, &comp_pool_lock);
|
2017-07-25 05:07:15 -04:00
|
|
|
if (zlib_pool_window == NULL)
|
|
|
|
|
zlib_pool_window = create_pool("zlib_window", size * items, MEM_F_SHARED);
|
2017-11-07 04:42:54 -05:00
|
|
|
HA_SPIN_UNLOCK(COMP_POOL_LOCK, &comp_pool_lock);
|
2017-07-25 05:07:15 -04:00
|
|
|
}
|
2014-12-24 12:07:55 -05:00
|
|
|
pool = zlib_pool_window;
|
2017-11-24 11:34:44 -05:00
|
|
|
ctx->zlib_window = buf = pool_alloc(pool);
|
2012-10-30 09:30:39 -04:00
|
|
|
break;
|
|
|
|
|
|
|
|
|
|
case 2:
|
2017-07-25 05:07:15 -04:00
|
|
|
if (zlib_pool_prev == NULL) {
|
2017-11-07 04:42:54 -05:00
|
|
|
HA_SPIN_LOCK(COMP_POOL_LOCK, &comp_pool_lock);
|
2017-07-25 05:07:15 -04:00
|
|
|
if (zlib_pool_prev == NULL)
|
|
|
|
|
zlib_pool_prev = create_pool("zlib_prev", size * items, MEM_F_SHARED);
|
2017-11-07 04:42:54 -05:00
|
|
|
HA_SPIN_UNLOCK(COMP_POOL_LOCK, &comp_pool_lock);
|
2017-07-25 05:07:15 -04:00
|
|
|
}
|
2014-12-24 12:07:55 -05:00
|
|
|
pool = zlib_pool_prev;
|
2017-11-24 11:34:44 -05:00
|
|
|
ctx->zlib_prev = buf = pool_alloc(pool);
|
2012-10-30 09:30:39 -04:00
|
|
|
break;
|
|
|
|
|
|
|
|
|
|
case 3:
|
2017-07-25 05:07:15 -04:00
|
|
|
if (zlib_pool_head == NULL) {
|
2017-11-07 04:42:54 -05:00
|
|
|
HA_SPIN_LOCK(COMP_POOL_LOCK, &comp_pool_lock);
|
2017-07-25 05:07:15 -04:00
|
|
|
if (zlib_pool_head == NULL)
|
|
|
|
|
zlib_pool_head = create_pool("zlib_head", size * items, MEM_F_SHARED);
|
2017-11-07 04:42:54 -05:00
|
|
|
HA_SPIN_UNLOCK(COMP_POOL_LOCK, &comp_pool_lock);
|
2017-07-25 05:07:15 -04:00
|
|
|
}
|
2014-12-24 12:07:55 -05:00
|
|
|
pool = zlib_pool_head;
|
2017-11-24 11:34:44 -05:00
|
|
|
ctx->zlib_head = buf = pool_alloc(pool);
|
2012-10-30 09:30:39 -04:00
|
|
|
break;
|
|
|
|
|
|
|
|
|
|
case 4:
|
2017-07-25 05:07:15 -04:00
|
|
|
if (zlib_pool_pending_buf == NULL) {
|
2017-11-07 04:42:54 -05:00
|
|
|
HA_SPIN_LOCK(COMP_POOL_LOCK, &comp_pool_lock);
|
2017-07-25 05:07:15 -04:00
|
|
|
if (zlib_pool_pending_buf == NULL)
|
|
|
|
|
zlib_pool_pending_buf = create_pool("zlib_pending_buf", size * items, MEM_F_SHARED);
|
2017-11-07 04:42:54 -05:00
|
|
|
HA_SPIN_UNLOCK(COMP_POOL_LOCK, &comp_pool_lock);
|
2017-07-25 05:07:15 -04:00
|
|
|
}
|
2014-12-24 12:07:55 -05:00
|
|
|
pool = zlib_pool_pending_buf;
|
2017-11-24 11:34:44 -05:00
|
|
|
ctx->zlib_pending_buf = buf = pool_alloc(pool);
|
2012-10-30 09:30:39 -04:00
|
|
|
break;
|
|
|
|
|
}
|
2012-11-20 05:25:20 -05:00
|
|
|
if (buf != NULL)
|
2017-07-25 05:07:15 -04:00
|
|
|
HA_ATOMIC_ADD(&zlib_used_memory, pool->size);
|
2012-11-07 10:12:57 -05:00
|
|
|
|
|
|
|
|
end:
|
2012-10-30 09:30:39 -04:00
|
|
|
|
2012-11-15 08:57:56 -05:00
|
|
|
/* deflateInit2() first allocates and checks the deflate_state, then if
|
|
|
|
|
* it succeeds, it allocates all other 4 areas at ones and checks them
|
|
|
|
|
* at the end. So we want to correctly count the rounds depending on when
|
|
|
|
|
* zlib is supposed to abort.
|
|
|
|
|
*/
|
|
|
|
|
if (buf || round)
|
|
|
|
|
round = (round + 1) % 5;
|
2012-10-30 09:30:39 -04:00
|
|
|
return buf;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static void free_zlib(void *opaque, void *ptr)
|
|
|
|
|
{
|
|
|
|
|
struct comp_ctx *ctx = opaque;
|
2012-11-10 11:49:37 -05:00
|
|
|
struct pool_head *pool = NULL;
|
2012-10-30 09:30:39 -04:00
|
|
|
|
|
|
|
|
if (ptr == ctx->zlib_window)
|
2012-11-07 10:12:57 -05:00
|
|
|
pool = zlib_pool_window;
|
2012-10-30 09:30:39 -04:00
|
|
|
else if (ptr == ctx->zlib_deflate_state)
|
2012-11-07 10:12:57 -05:00
|
|
|
pool = zlib_pool_deflate_state;
|
2012-10-30 09:30:39 -04:00
|
|
|
else if (ptr == ctx->zlib_prev)
|
2012-11-07 10:12:57 -05:00
|
|
|
pool = zlib_pool_prev;
|
2012-10-30 09:30:39 -04:00
|
|
|
else if (ptr == ctx->zlib_head)
|
2012-11-07 10:12:57 -05:00
|
|
|
pool = zlib_pool_head;
|
2012-10-30 09:30:39 -04:00
|
|
|
else if (ptr == ctx->zlib_pending_buf)
|
2012-11-07 10:12:57 -05:00
|
|
|
pool = zlib_pool_pending_buf;
|
2012-10-30 09:30:39 -04:00
|
|
|
|
2017-11-24 11:34:44 -05:00
|
|
|
pool_free(pool, ptr);
|
2017-07-25 05:07:15 -04:00
|
|
|
HA_ATOMIC_SUB(&zlib_used_memory, pool->size);
|
2012-10-30 09:30:39 -04:00
|
|
|
}
|
|
|
|
|
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
/**************************
|
|
|
|
|
**** gzip algorithm ****
|
|
|
|
|
***************************/
|
2015-03-28 10:46:00 -04:00
|
|
|
static int gzip_init(struct comp_ctx **comp_ctx, int level)
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
{
|
2012-11-16 12:06:41 -05:00
|
|
|
z_stream *strm;
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
|
2012-11-16 12:06:41 -05:00
|
|
|
if (init_comp_ctx(comp_ctx) < 0)
|
|
|
|
|
return -1;
|
2012-11-07 10:12:57 -05:00
|
|
|
|
2012-11-16 12:06:41 -05:00
|
|
|
strm = &(*comp_ctx)->strm;
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
|
2016-12-22 13:46:17 -05:00
|
|
|
if (deflateInit2(strm, level, Z_DEFLATED, global_tune_zlibwindowsize + 16, global_tune_zlibmemlevel, Z_DEFAULT_STRATEGY) != Z_OK) {
|
2012-11-16 12:06:41 -05:00
|
|
|
deinit_comp_ctx(comp_ctx);
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
return -1;
|
2012-11-16 12:06:41 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
(*comp_ctx)->cur_lvl = level;
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
|
}
|
2015-03-28 12:00:39 -04:00
|
|
|
|
|
|
|
|
/* Raw deflate algorithm */
|
|
|
|
|
static int raw_def_init(struct comp_ctx **comp_ctx, int level)
|
|
|
|
|
{
|
|
|
|
|
z_stream *strm;
|
|
|
|
|
|
|
|
|
|
if (init_comp_ctx(comp_ctx) < 0)
|
|
|
|
|
return -1;
|
|
|
|
|
|
|
|
|
|
strm = &(*comp_ctx)->strm;
|
|
|
|
|
|
2016-12-22 13:46:17 -05:00
|
|
|
if (deflateInit2(strm, level, Z_DEFLATED, -global_tune_zlibwindowsize, global_tune_zlibmemlevel, Z_DEFAULT_STRATEGY) != Z_OK) {
|
2015-03-28 12:00:39 -04:00
|
|
|
deinit_comp_ctx(comp_ctx);
|
|
|
|
|
return -1;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
(*comp_ctx)->cur_lvl = level;
|
|
|
|
|
return 0;
|
|
|
|
|
}
|
|
|
|
|
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
/**************************
|
|
|
|
|
**** Deflate algorithm ****
|
|
|
|
|
***************************/
|
|
|
|
|
|
2015-03-28 10:46:00 -04:00
|
|
|
static int deflate_init(struct comp_ctx **comp_ctx, int level)
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
{
|
2012-11-16 12:06:41 -05:00
|
|
|
z_stream *strm;
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
|
2012-11-16 12:06:41 -05:00
|
|
|
if (init_comp_ctx(comp_ctx) < 0)
|
|
|
|
|
return -1;
|
|
|
|
|
|
|
|
|
|
strm = &(*comp_ctx)->strm;
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
|
2016-12-22 13:46:17 -05:00
|
|
|
if (deflateInit2(strm, level, Z_DEFLATED, global_tune_zlibwindowsize, global_tune_zlibmemlevel, Z_DEFAULT_STRATEGY) != Z_OK) {
|
2012-11-16 12:06:41 -05:00
|
|
|
deinit_comp_ctx(comp_ctx);
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
return -1;
|
2012-11-16 12:06:41 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
(*comp_ctx)->cur_lvl = level;
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
|
}
|
|
|
|
|
|
2012-11-19 06:35:37 -05:00
|
|
|
/* Return the size of consumed data or -1 */
|
2015-03-28 10:46:00 -04:00
|
|
|
static int deflate_add_data(struct comp_ctx *comp_ctx, const char *in_data, int in_len, struct buffer *out)
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
{
|
|
|
|
|
int ret;
|
2012-11-19 06:35:37 -05:00
|
|
|
z_stream *strm = &comp_ctx->strm;
|
2018-06-07 12:46:28 -04:00
|
|
|
char *out_data = b_tail(out);
|
2018-06-15 07:59:36 -04:00
|
|
|
int out_len = b_room(out);
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
|
|
|
|
|
if (in_len <= 0)
|
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
if (out_len <= 0)
|
|
|
|
|
return -1;
|
|
|
|
|
|
|
|
|
|
strm->next_in = (unsigned char *)in_data;
|
|
|
|
|
strm->avail_in = in_len;
|
|
|
|
|
strm->next_out = (unsigned char *)out_data;
|
|
|
|
|
strm->avail_out = out_len;
|
|
|
|
|
|
|
|
|
|
ret = deflate(strm, Z_NO_FLUSH);
|
|
|
|
|
if (ret != Z_OK)
|
|
|
|
|
return -1;
|
|
|
|
|
|
|
|
|
|
/* deflate update the available data out */
|
2018-06-28 12:17:23 -04:00
|
|
|
b_add(out, out_len - strm->avail_out);
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
|
2012-11-19 06:35:37 -05:00
|
|
|
return in_len - strm->avail_in;
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
}
|
|
|
|
|
|
2015-03-28 14:17:31 -04:00
|
|
|
static int deflate_flush_or_finish(struct comp_ctx *comp_ctx, struct buffer *out, int flag)
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
{
|
|
|
|
|
int ret;
|
|
|
|
|
int out_len = 0;
|
2012-10-30 10:52:53 -04:00
|
|
|
z_stream *strm = &comp_ctx->strm;
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
|
BUG/MAJOR: compression: initialize avail_in/next_in even during flush
For quite some time, a few users have been experiencing random crashes
when compressing with zlib, from versions 1.2.3 to 1.2.8 included.
Upon thourough investigation in zlib's deflate.c, it appeared obvious
that avail_in and next_in are used during the flush operation and need
to be initialized, while admittedly it's not obvious in the documentation.
By simply forcing both values to -1 it's possible to immediately reproduce
the exact crash that these users have been experiencing :
(gdb) bt
#0 0x00007fa73ce10c00 in __memcpy_sse2 () from /lib64/libc.so.6
#1 0x00007fa73e0c5d49 in ?? () from /lib64/libz.so.1
#2 0x00007fa73e0c68e0 in ?? () from /lib64/libz.so.1
#3 0x00007fa73e0c73c7 in deflate () from /lib64/libz.so.1
#4 0x00000000004dca1c in deflate_flush_or_finish (comp_ctx=0x7b6580, out=0x7fa73e5bd010, flag=2) at src/compression.c:808
#5 0x00000000004dcb60 in deflate_flush (comp_ctx=0x7b6580, out=0x7fa73e5bd010) at src/compression.c:835
#6 0x00000000004dbc50 in http_compression_buffer_end (s=0x7c0050, in=0x7c00a8, out=0x78adf0 <tmpbuf.24662>, end=0) at src/compression.c:249
#7 0x000000000048bb5f in http_response_forward_body (s=0x7c0050, res=0x7c00a0, an_bit=1048576) at src/proto_http.c:7173
#8 0x00000000004cce54 in process_stream (t=0x7bffd8) at src/stream.c:1939
#9 0x0000000000427ddf in process_runnable_tasks () at src/task.c:238
#10 0x0000000000419892 in run_poll_loop () at src/haproxy.c:1573
#11 0x000000000041a4a5 in main (argc=4, argv=0x7fffcda38348) at src/haproxy.c:1933
Note that for all reports deflate_flush_or_finish() was always involved.
The crash is very hard to reproduce when using regular traffic because it
requires that the combination of avail_in and next_in are inadequate so
that the memcpy() call reads out of bounds. But this can very likely
happen when the input buffer points to an area reused by another stream
when the flush has been interrupted due to a full output buffer. This
also explains why this report is recent, as dynamic buffer allocation
was introduced in 1.6.
Anyway it's not acceptable to call a function with a randomly set input
buffer. The deflate() function explicitly checks for the case where both
avail_in and next_in are null and doesn't use it in this case during a
flush, so this is the best solution.
Special thanks to Sasha Litvak, James Hartshorn and Paul Bauer for
reporting very useful stack traces which were critical to finding the
root cause of this bug.
This fix must be backported into 1.6 and 1.5, though 1.5 is less likely to
trigger this case given that it keeps its own buffers allocated all along
the session's life.
2016-08-08 10:41:01 -04:00
|
|
|
strm->next_in = NULL;
|
|
|
|
|
strm->avail_in = 0;
|
2018-06-07 12:46:28 -04:00
|
|
|
strm->next_out = (unsigned char *)b_tail(out);
|
2018-06-15 07:59:36 -04:00
|
|
|
strm->avail_out = b_room(out);
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
|
|
|
|
|
ret = deflate(strm, flag);
|
|
|
|
|
if (ret != Z_OK && ret != Z_STREAM_END)
|
|
|
|
|
return -1;
|
|
|
|
|
|
2018-06-15 07:59:36 -04:00
|
|
|
out_len = b_room(out) - strm->avail_out;
|
2018-06-28 12:17:23 -04:00
|
|
|
b_add(out, out_len);
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
|
2012-11-20 11:01:01 -05:00
|
|
|
/* compression limit */
|
|
|
|
|
if ((global.comp_rate_lim > 0 && (read_freq_ctr(&global.comp_bps_out) > global.comp_rate_lim)) || /* rate */
|
|
|
|
|
(idle_pct < compress_min_idle)) { /* idle */
|
|
|
|
|
/* decrease level */
|
|
|
|
|
if (comp_ctx->cur_lvl > 0) {
|
|
|
|
|
comp_ctx->cur_lvl--;
|
2012-11-09 11:05:39 -05:00
|
|
|
deflateParams(&comp_ctx->strm, comp_ctx->cur_lvl, Z_DEFAULT_STRATEGY);
|
|
|
|
|
}
|
2012-11-20 11:01:01 -05:00
|
|
|
|
|
|
|
|
} else if (comp_ctx->cur_lvl < global.tune.comp_maxlevel) {
|
|
|
|
|
/* increase level */
|
|
|
|
|
comp_ctx->cur_lvl++ ;
|
|
|
|
|
deflateParams(&comp_ctx->strm, comp_ctx->cur_lvl, Z_DEFAULT_STRATEGY);
|
2012-11-09 11:05:39 -05:00
|
|
|
}
|
|
|
|
|
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
return out_len;
|
|
|
|
|
}
|
|
|
|
|
|
2015-03-28 14:17:31 -04:00
|
|
|
static int deflate_flush(struct comp_ctx *comp_ctx, struct buffer *out)
|
|
|
|
|
{
|
|
|
|
|
return deflate_flush_or_finish(comp_ctx, out, Z_SYNC_FLUSH);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static int deflate_finish(struct comp_ctx *comp_ctx, struct buffer *out)
|
|
|
|
|
{
|
|
|
|
|
return deflate_flush_or_finish(comp_ctx, out, Z_FINISH);
|
|
|
|
|
}
|
|
|
|
|
|
2015-03-28 10:46:00 -04:00
|
|
|
static int deflate_end(struct comp_ctx **comp_ctx)
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
{
|
2012-11-16 12:06:41 -05:00
|
|
|
z_stream *strm = &(*comp_ctx)->strm;
|
|
|
|
|
int ret;
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
|
2012-11-16 12:06:41 -05:00
|
|
|
ret = deflateEnd(strm);
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
|
2012-11-16 12:06:41 -05:00
|
|
|
deinit_comp_ctx(comp_ctx);
|
|
|
|
|
|
|
|
|
|
return ret;
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
}
|
|
|
|
|
|
2016-12-22 13:46:17 -05:00
|
|
|
/* config parser for global "tune.zlibmemlevel" */
|
|
|
|
|
static int zlib_parse_global_memlevel(char **args, int section_type, struct proxy *curpx,
|
|
|
|
|
struct proxy *defpx, const char *file, int line,
|
|
|
|
|
char **err)
|
|
|
|
|
{
|
|
|
|
|
if (too_many_args(1, args, err, NULL))
|
|
|
|
|
return -1;
|
|
|
|
|
|
|
|
|
|
if (*(args[1]) == 0) {
|
|
|
|
|
memprintf(err, "'%s' expects a numeric value between 1 and 9.", args[0]);
|
|
|
|
|
return -1;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
global_tune_zlibmemlevel = atoi(args[1]);
|
|
|
|
|
if (global_tune_zlibmemlevel < 1 || global_tune_zlibmemlevel > 9) {
|
|
|
|
|
memprintf(err, "'%s' expects a numeric value between 1 and 9.", args[0]);
|
|
|
|
|
return -1;
|
|
|
|
|
}
|
|
|
|
|
return 0;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
/* config parser for global "tune.zlibwindowsize" */
|
|
|
|
|
static int zlib_parse_global_windowsize(char **args, int section_type, struct proxy *curpx,
|
|
|
|
|
struct proxy *defpx, const char *file, int line,
|
|
|
|
|
char **err)
|
|
|
|
|
{
|
|
|
|
|
if (too_many_args(1, args, err, NULL))
|
|
|
|
|
return -1;
|
|
|
|
|
|
|
|
|
|
if (*(args[1]) == 0) {
|
|
|
|
|
memprintf(err, "'%s' expects a numeric value between 8 and 15.", args[0]);
|
|
|
|
|
return -1;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
global_tune_zlibwindowsize = atoi(args[1]);
|
|
|
|
|
if (global_tune_zlibwindowsize < 8 || global_tune_zlibwindowsize > 15) {
|
|
|
|
|
memprintf(err, "'%s' expects a numeric value between 8 and 15.", args[0]);
|
|
|
|
|
return -1;
|
|
|
|
|
}
|
|
|
|
|
return 0;
|
|
|
|
|
}
|
|
|
|
|
|
MEDIUM: HTTP compression (zlib library support)
This commit introduces HTTP compression using the zlib library.
http_response_forward_body has been modified to call the compression
functions.
This feature includes 3 algorithms: identity, gzip and deflate:
* identity: this is mostly for debugging, and it was useful for
developping the compression feature. With Content-Length in input, it
is making each chunk with the data available in the current buffer.
With chunks in input, it is rechunking, the output chunks will be
bigger or smaller depending of the size of the input chunk and the
size of the buffer. Identity does not apply any change on data.
* gzip: same as identity, but applying a gzip compression. The data
are deflated using the Z_NO_FLUSH flag in zlib. When there is no more
data in the input buffer, it flushes the data in the output buffer
(Z_SYNC_FLUSH). At the end of data, when it receives the last chunk in
input, or when there is no more data to read, it writes the end of
data with Z_FINISH and the ending chunk.
* deflate: same as gzip, but with deflate algorithm and zlib format.
Note that this algorithm has ambiguous support on many browsers and
no support at all from recent ones. It is strongly recommended not
to use it for anything else than experimentation.
You can't choose the compression ratio at the moment, it will be set to
Z_BEST_SPEED (1), as tests have shown very little benefit in terms of
compression ration when going above for HTML contents, at the cost of
a massive CPU impact.
Compression will be activated depending of the Accept-Encoding request
header. With identity, it does not take care of that header.
To build HAProxy with zlib support, use USE_ZLIB=1 in the make
parameters.
This work was initially started by David Du Colombier at Exceliance.
2012-10-23 04:25:10 -04:00
|
|
|
#endif /* USE_ZLIB */
|
|
|
|
|
|
2016-12-22 13:46:17 -05:00
|
|
|
|
|
|
|
|
/* config keyword parsers */
|
|
|
|
|
static struct cfg_kw_list cfg_kws = {ILH, {
|
|
|
|
|
#ifdef USE_ZLIB
|
|
|
|
|
{ CFG_GLOBAL, "tune.zlib.memlevel", zlib_parse_global_memlevel },
|
|
|
|
|
{ CFG_GLOBAL, "tune.zlib.windowsize", zlib_parse_global_windowsize },
|
|
|
|
|
#endif
|
|
|
|
|
{ 0, NULL, NULL }
|
|
|
|
|
}};
|
|
|
|
|
|
2018-11-25 13:14:37 -05:00
|
|
|
INITCALL1(STG_REGISTER, cfg_register_keywords, &cfg_kws);
|
|
|
|
|
|
2013-04-20 11:33:20 -04:00
|
|
|
__attribute__((constructor))
|
|
|
|
|
static void __comp_fetch_init(void)
|
|
|
|
|
{
|
MAJOR: compression: integrate support for libslz
This library is designed to emit a zlib-compatible stream with no
memory usage and to favor resource savings over compression ratio.
While zlib requires 256 kB of RAM per compression context (and can only
support 4000 connections per GB of RAM), the stateless compression
offered by libslz does not need to retain buffers between subsequent
calls. In theory this slightly reduces the compression ratio but in
practice it does not have that much of an effect since the zlib
window is limited to 32kB.
Libslz is available at :
http://git.1wt.eu/web?p=libslz.git
It was designed for web compression and provides a lot of savings
over zlib in haproxy. Here are the preliminary results on a single
core of a core2-quad 3.0 GHz in 32-bit for only 300 concurrent
sessions visiting the home page of www.haproxy.org (76 kB) with
the default 16kB buffers :
BW In BW Out BW Saved Ratio memory VSZ/RSS
zlib 237 Mbps 92 Mbps 145 Mbps 2.58 84M / 69M
slz 733 Mbps 380 Mbps 353 Mbps 1.93 5.9M / 4.2M
So while the compression ratio is lower, the bandwidth savings are
much more important due to the significantly lower compression cost
which allows to consume even more data from the servers. In the
example above, zlib became the bottleneck at 24% of the output
bandwidth. Also the difference in memory usage is obvious.
More tests run on a single core of a core i5-3320M, with 500 concurrent
users and the default 16kB buffers :
At 100% CPU (no limit) :
BW In BW Out BW Saved Ratio memory VSZ/RSS hits/s
zlib 480 Mbps 188 Mbps 292 Mbps 2.55 130M / 101M 744
slz 1700 Mbps 810 Mbps 890 Mbps 2.10 23.7M / 9.7M 2382
At 85% CPU (limited) :
BW In BW Out BW Saved Ratio memory VSZ/RSS hits/s
zlib 1240 Mbps 976 Mbps 264 Mbps 1.27 130M / 100M 1738
slz 1600 Mbps 976 Mbps 624 Mbps 1.64 23.7M / 9.7M 2210
The most important benefit really happens when the CPU usage is
limited by "maxcompcpuusage" or the BW limited by "maxcomprate" :
in order to preserve resources, haproxy throttles the compression
ratio until usage is within limits. Since slz is much cheaper, the
average compression ratio is much higher and the input bandwidth
is quite higher for one Gbps output.
Other tests made with some reference files :
BW In BW Out BW Saved Ratio hits/s
daniels.html zlib 1320 Mbps 163 Mbps 1157 Mbps 8.10 1925
slz 3600 Mbps 580 Mbps 3020 Mbps 6.20 5300
tv.com/listing zlib 980 Mbps 124 Mbps 856 Mbps 7.90 310
slz 3300 Mbps 553 Mbps 2747 Mbps 5.97 1100
jquery.min.js zlib 430 Mbps 180 Mbps 250 Mbps 2.39 547
slz 1470 Mbps 764 Mbps 706 Mbps 1.92 1815
bootstrap.min.css zlib 790 Mbps 165 Mbps 625 Mbps 4.79 777
slz 2450 Mbps 650 Mbps 1800 Mbps 3.77 2400
So on top of saving a lot of memory, slz is constantly 2.5-3.5 times
faster than zlib and results in providing more savings for a fixed CPU
usage. For links smaller than 100 Mbps, zlib still provides a better
compression ratio, at the expense of a much higher CPU usage.
Larger input files provide slightly higher bandwidth for both libs, at
the expense of a bit more memory usage for zlib (it converges to 256kB
per connection).
2015-03-28 21:32:06 -04:00
|
|
|
#ifdef USE_SLZ
|
|
|
|
|
slz_make_crc_table();
|
|
|
|
|
slz_prepare_dist_table();
|
|
|
|
|
#endif
|
2018-11-26 08:44:03 -05:00
|
|
|
|
2016-12-22 13:46:17 -05:00
|
|
|
#if defined(USE_ZLIB) && defined(DEFAULT_MAXZLIBMEM)
|
2018-11-26 04:24:45 -05:00
|
|
|
global.maxzlibmem = DEFAULT_MAXZLIBMEM * 1024U * 1024U;
|
2016-12-22 13:46:17 -05:00
|
|
|
#endif
|
2018-11-26 04:19:54 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static void comp_register_build_opts(void)
|
|
|
|
|
{
|
|
|
|
|
char *ptr = NULL;
|
|
|
|
|
int i;
|
|
|
|
|
|
2016-12-21 13:30:30 -05:00
|
|
|
#ifdef USE_ZLIB
|
|
|
|
|
memprintf(&ptr, "Built with zlib version : " ZLIB_VERSION);
|
|
|
|
|
memprintf(&ptr, "%s\nRunning on zlib version : %s", ptr, zlibVersion());
|
|
|
|
|
#elif defined(USE_SLZ)
|
|
|
|
|
memprintf(&ptr, "Built with libslz for stateless compression.");
|
2017-01-11 09:24:35 -05:00
|
|
|
#else
|
|
|
|
|
memprintf(&ptr, "Built without compression support (neither USE_ZLIB nor USE_SLZ are set).");
|
2016-12-21 13:30:30 -05:00
|
|
|
#endif
|
|
|
|
|
memprintf(&ptr, "%s\nCompression algorithms supported :", ptr);
|
|
|
|
|
|
|
|
|
|
for (i = 0; comp_algos[i].cfg_name; i++)
|
|
|
|
|
memprintf(&ptr, "%s%s %s(\"%s\")", ptr, (i == 0 ? "" : ","), comp_algos[i].cfg_name, comp_algos[i].ua_name);
|
|
|
|
|
|
|
|
|
|
if (i == 0)
|
|
|
|
|
memprintf(&ptr, "%s none", ptr);
|
|
|
|
|
|
|
|
|
|
hap_register_build_opts(ptr, 1);
|
2013-04-20 11:33:20 -04:00
|
|
|
}
|
2018-11-26 04:19:54 -05:00
|
|
|
|
|
|
|
|
INITCALL0(STG_REGISTER, comp_register_build_opts);
|
