httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Brian Behlendorf <br...@hyperreal.org>
Subject Re: process model again...
Date Wed, 08 Apr 1998 07:54:59 GMT
At 12:39 AM 4/8/98 -0700, Dean Gaudet wrote:
>There has to be a better algorithm.  I don't like people having to tweak
>these numbers.
>
>I'm thinking of something that actually is based on the rate of
>connections, and the average service time.  Anyone study queuing theory? 
>I avoided that unfortunately. 

No :(  I still think that a memory-based approach would be better.  The
worst thing in the world to do is to go to swap; it's better to leave a
connection waiting than accept it and swap yourself or some other process
out.  The flip side is that most people are running dedicated web servers
these days, and even though they have to "play nice" with the other
software running there, for the most part there's no reason not to leave
lots of children around so long as they don't cause swapping.

On a slightly less radical level; what about never killing children?  Let
them die off thanks to MaxRequestsPerChild (and never let that be infinity
:).... if the OS really needs that memory it can swap those children out.

	Brian


--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=--
"Optimism is a strategy for making                         brian@apache.org
a better future." - Noam Chomsky                        brian@hyperreal.org

Mime
View raw message