[IRC-DEV] Diferencias de CPU entre kqueue() y select()

Toni García zoltan at teleline.es
Mon Dec 30 09:40:05 CET 2002


Muy interesante :)

> -----Mensaje original-----
> De: owner-dalnet-src at DAL.NET 
> [mailto:owner-dalnet-src at DAL.NET] En nombre de Perry Lorier
> Enviado el: lunes, 30 de diciembre de 2002 2:36
> Para: David Schwartz
> CC: Lucas Madar; dalnet-src at DAL.NET
> Asunto: Re: DALnet-src: Technical Discussion: HTM and reading 
> from clients
> Importancia: Alta
> 
> 
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> > 	But this is one of the places where you can really 
> reduce your system CPU 
> > consumption. What's the ratio of user CPU time to system 
> CPU time on a 
> > typical DALnet server these days? If system CPU time is 40% 
> or more of what 
> > user CPU time is, I'd focus on the parameters you pass to 
> select/poll and 
> > what steps you take to:
> 
> Undernet moved from poll() and select() to kqueue(), 
> /dev/poll, and all
> the rest.
> 
> To show the effect of this on a server, heres the CPU load of a server
> that holds about 4k clients:
> 
http://www.break.net/ircu10-to-11.png

On the left is kqueue(), on the right is select().

Almost all an IRC servers time it spent doing select()/poll() and the
associated housekeeping.

- -- 
Excellent day for putting Slinkies on an escalator.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.1 (GNU/Linux)
Comment: Only when you are sure they have you, can you stop being
paranoid

iD8DBQE+D6L9cAgRpy8z8UQRAhZVAJ9bj4LNHYxWH5xklDLVZgrCw1gA8gCgqq74
nt+kwLzmXTikOeJEOT8xhhQ=
=/EBL
-----END PGP SIGNATURE-----
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: Re_ DALnet-src_ Technical Discussion_ HTM and reading from clients.txt
URL: <http://mailman.jcea.es/pipermail/irc-dev/attachments/20021230/08750a9f/attachment.txt>


More information about the IRC-Dev mailing list