mirror of
http://git.haproxy.org/git/haproxy.git/
synced 2024-12-14 07:24:32 +00:00
haproxy public development tree
5bd8c376ad
This code provides support for linux 2.6 kernel splicing. This feature appeared in kernel 2.6.25, but initial implementations were awkward and buggy. A kernel >= 2.6.29-rc1 is recommended, as well as some optimization patches. Using pipes, this code is able to pass network data directly between sockets. The pipes are a bit annoying to manage (fd creation, release, ...) but finally work quite well. Preliminary tests show that on high bandwidths, there's a substantial gain (approx +50%, only +20% with kernel workarounds for corruption bugs). With 2000 concurrent connections, with Myricom NICs, haproxy now more easily achieves 4.5 Gbps for 1 process and 6 Gbps for two processes buffers. 8-9 Gbps are easily reached with smaller numbers of connections. We also try to splice out immediately after a splice in by making profit from the new ability for a data producer to notify the consumer that data are available. Doing this ensures that the data are immediately transferred between sockets without latency, and without having to re-poll. Performance on small packets has considerably increased due to this method. Earlier kernels return only one TCP segment at a time in non-blocking splice-in mode, while newer return as many segments as may fit in the pipe. To work around this limitation without hurting more recent kernels, we try to collect as much data as possible, but we stop when we believe we have read 16 segments, then we forward everything at once. It also ensures that even upon shutdown or EAGAIN the data will be forwarded. Some tricks were necessary because the splice() syscall does not make a difference between missing data and a pipe full, it always returns EAGAIN. The trick consists in stop polling in case of EAGAIN and a non empty pipe. The receiver waits for the buffer to be empty before using the pipe. This is in order to avoid confusion between buffer data and pipe data. The BF_EMPTY flag now covers the pipe too. Right now the code is disabled by default. It needs to be built with CONFIG_HAP_LINUX_SPLICE, and the instances intented to use splice() must have "option splice-response" (or option splice-request) enabled. It is probably desirable to keep a pool of pre-allocated pipes to avoid having to create them for every session. This will be worked on later. Preliminary tests show very good results, even with the kernel workaround causing one memcpy(). At 3000 connections, performance has moved from 3.2 Gbps to 4.7 Gbps. |
||
|---|---|---|
| contrib/netsnmp-perl | ||
| doc | ||
| examples | ||
| include | ||
| src | ||
| tests | ||
| .gitignore | ||
| CHANGELOG | ||
| CONTRIB | ||
| LICENSE | ||
| Makefile | ||
| Makefile.bsd | ||
| Makefile.osx | ||
| README | ||
| ROADMAP | ||
| SUBVERS | ||
| TODO | ||
| VERDATE | ||
| VERSION | ||
-------------------
H A - P r o x y
How to build it
-------------------
version 1.3.15
willy tarreau
2008/05/25
To build haproxy, you will need :
- GNU make. Neither Solaris nor OpenBSD's make work with this makefile.
However, specific Makefiles for BSD and OSX are provided.
- GCC between 2.91 and 4.3. Others may work, but not tested.
- GNU ld
Also, you might want to build with libpcre support, which will provide a very
efficient regex implementation and will also fix some badness on Solaris's one.
To build haproxy, you have to choose your target OS amongst the following ones
and assign it to the TARGET variable :
- linux22 for Linux 2.2
- linux24 for Linux 2.4 and above (default)
- linux24e for Linux 2.4 with support for a working epoll (> 0.21)
- linux24eold for Linux 2.4 with support for a broken epoll (<= 0.21)
- linux26 for Linux 2.6 and above
- solaris for Solaris 8 or 10 (others untested)
- freebsd for FreeBSD 5 to 6.2 (others untested)
- openbsd for OpenBSD 3.1 to 3.7 (others untested)
- generic for any other OS.
- custom to manually adjust every setting
You may also choose your CPU to benefit from some optimizations. This is
particularly important on UltraSparc machines. For this, you can assign
one of the following choices to the CPU variable :
- i686 for intel PentiumPro, Pentium 2 and above, AMD Athlon
- i586 for intel Pentium, AMD K6, VIA C3.
- ultrasparc : Sun UltraSparc I/II/III/IV processor
- generic : any other processor or no specific optimization. (default)
Alternatively, you may just set the CPU_CFLAGS value to the optimal GCC options
for your platform.
If your system supports PCRE (Perl Compatible Regular Expressions), then you
really should build with libpcre which is between 2 and 10 times faster than
other libc implementations. Regex are used for header processing (deletion,
rewriting, allow, deny). The only inconvenient of libpcre is that it is not
yet widely spread, so if you build for other systems, you might get into
trouble if they don't have the dynamic library. In this situation, you should
statically link libpcre into haproxy so that it will not be necessary to
install it on target systems. Available build options for PCRE are :
- USE_PCRE=1 to use libpcre, in whatever form is available on your system
(shared or static)
- USE_STATIC_PCRE=1 to use a static version of libpcre even if the dynamic
one is available. This will enhance portability.
- with no option, use your OS libc's standard regex implemntation (default).
Warning! group references on Solaris seem broken. Use static-pcre whenever
possible.
By default, the DEBUG variable is set to '-g' to enable debug symbols. It is
not wise to disable it on uncommon systems, because it's often the only way to
get a complete core when you need one. Otherwise, you can set DEBUG to '-s' to
strip the binary.
For example, I use this to build for Solaris 8 :
$ make TARGET=solaris CPU=ultrasparc USE_STATIC_PCRE=1
And I build it this way on OpenBSD or FreeBSD :
$ make -f Makefile.bsd REGEX=pcre DEBUG= COPTS.generic="-Os -fomit-frame-pointer -mgnu"
If you need to pass other defines, includes, libraries, etc... then please
check the Makefile to see which ones will be available in your case, and
use the USE_* variables in the GNU Makefile, or ADDINC, ADDLIB, and DEFINE
variables in the BSD makefiles.
-- end