Re: Process limits for epoll tests

Davide Libenzi (davidel@xmailserver.org)
Tue, 1 Apr 2003 08:15:04 -0800 (PST)


On Tue, 1 Apr 2003, Philippe Meloche (LMC) wrote:

> Hi,

Hi,

> We tried to reproduce the tests you've done with /dev/epoll and we've
> come to ask us some questions.

I hope you are not using the old /dev/epoll since it is no more supported.
The kernel 2.5.x has epoll, that is supported :

http://www.xmailserver.org/linux-patches/epoll.txt
http://www.xmailserver.org/linux-patches/epoll_create.txt
http://www.xmailserver.org/linux-patches/epoll_ctl.txt
http://www.xmailserver.org/linux-patches/epoll_wait.txt

> 1. How did you managed httperf to perform more than 1024 connections
> when it's using select() ?

Yes, you can't have more than 1024 simultaneous connections with httperf.
But if the session time is short, you can achieve a pretty high rate with
1024 maximum connections. Lately I am using the http blaster ( ver very
simple http loader ) that uses epoll. The latest source code is here :

http://www.xmailserver.org/linux-patches/epoll-lib-0.7.tar.gz

You need this coroutine library to build that package :

http://www.xmailserver.org/libpcl.html

> 2. Did you get some errors like client-timeout or connections reset when
> you were doing your tests ?
>
> 3. How much time did a burst test ( 27000 connections and 2 calls per
> connection ) last and how many sample did httperf
> took during those tests.

You definitely have to tweak kernel/net params to run the test ( port
range, fin timeout, tw recycle, max files, ... ).

- Davide

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/