TUX SpecWeb99 configuration (was: Re: tux3-2.6.8.1-A5 performance)

[Yaoping Ruan]

>I'm curious - which OS did you benchmark Flash on?

It is FC2 with 2.6.8.1 kernel

> If it was on Linux, which family of syscalls does it use for polling
(select/poll/epoll)?

It uses epoll. Actually we did an experiment before to measure the
effectiveness of polling. The conclusion is that the throughput doesn't
change too much if you have more than 20% connections doing "real" work
(active connections). I assume that the performance of kevent( ) on
FreeBSD is close to epoll( ). You may refer to:
http://www.cs.princeton.edu/~yruan/flexiclient/

>For the most, concurrency does not exceed a few dozen requests at the
same time. So I'd agree >with Kees' remark and suggest (if you want to
evaluate this further) that you
>try to benchmark with ab and vary the concurrency from a few
connections
>to a few thousand concurrent connections and observe the level of
>performance degradation compared to TUX.

During my tests, I used 1200 concurrent connections, also I tried 2400.
It would be useful to have SPECWeb99 tests and see what I am missing,
because the results are public available on SPEC's site and TUX is the
dominant server.

So, Could anybody let me know how to put SPECWeb99 dynamic API? From the
source code, CAD_u.tux.so and DYN_g.tux.cgi will do the right thing.
After I put them under Apache's doc dir, and have Apache run at port
8080, I couldn't get the right response. So my questions are:
1. Do I need Apache to run CAD_u.tux.so?
2. Does the API for RHCA 3.2 also work for recent TUX patches?

-Yaoping