Simple "hello world" benchmarks for some web servers and frameworks. Done with high influence of TechEmpower's FrameworkBenchmarks and some others, see Links.

Each web server/framework was configured to gain maximum RPS on the test stand (see below). For example disabled kernel-poll in Erlang gives several KRPS. Your mileage may vary.

Configurations with huge performance difference has their own benchmarks (e.g. warp-threads and warp-processes).

Compile and install required dependencies with:

$ make install-deps

You could run this and other steps only for single directory, for example:

$ make gunicorn-sync/install-deps

$ make

Check your ulimit and sysctl settings with:

$ make check

See aditional info about system tuning in Links section.

$ make bench

Results will be available inside the logs directory.

Similar benchmarks:
http://www.techempower.com/benchmarks/
https://github.com/TechEmpower/FrameworkBenchmarks
https://github.com/methane/echoserver
http://www.yesodweb.com/blog/2011/03/preliminary-warp-cross-language-benchmarks

System tuning:
http://gwan.com/en_apachebench_httperf.html
http://selectnull.leongkui.me/2013/06/27/be-careful-with-tcp_tw_recycle/
http://www.fromdual.com/huge-amount-of-time-wait-connections

Tools:
https://github.com/lighttpd/weighttp (uses epoll and threads, nice utility)
http://www.hpl.hp.com/research/linux/httperf/
https://httpd.apache.org/docs/2.2/programs/ab.html (don't bench for >50K RPS, it will simply overload single CPU core)

Other:
http://aosabook.org/en/posa/warp.html
http://haskell.cs.yale.edu/wp-content/uploads/2013/08/hask035-voellmy.pdf
http://mervine.net/performance-testing-with-httperf
https://www.mnot.net/blog/2011/05/18/http_benchmark_rules (Benchmarks from this repo misses half of the rules… Well, it can't be helped.)
http://www.cse.wustl.edu/ANCS/2007/slides/Bryan%20Veal%20ANCS%20Presentation.pdf

To get the most honest results you really should prepare and run all these benchmarks by yourself. And tune some of them if necessary.

For example while doing my own behchmarks I discover some strange "bugs" in others:

• Warp's performance in Round 9 of TechEmpower's benchmarks seems weird: it even slower than Python!
• NodeJS in Snoyman's bench shouldn't be that slow: after some investigations you will find that node have been running only on single CPU core

So you should never believe numbers from the net and check everything by yourself. Don't believe mine as well. But for curious one and as a memo I've recorded them anyway.

Test stand:

• Gentoo x86-64 (all software was built from sources with appropriate GCC flags)
• Linux 3.13
• Intel i7 3820 3.6 GHz (4 physical cores, 8 virtual cores with HT)
• 32GB RAM
• weighttp -c 1000 -t 10 -k
• Web server and test clients at the same host, connected via loopback

Platform versions:

• Python 2.7.5
• NodeJS 0.10.26
• GHC 7.8.2
• Erlang 17.0 (+hipe +kpoll +smp)
• Go 1.2

Software versions:

• Latest available in git master or in standart language's package manager at the moment of testing

Check logs directory for resulting logs.