httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Dean Gaudet <dgau...@arctic.org>
Subject Re: [PATCH] incomplete: scoreboard work
Date Tue, 26 Aug 1997 19:31:40 GMT


On Tue, 26 Aug 1997, Brian Behlendorf wrote:

> At 01:47 AM 8/26/97 -0700, Dean Gaudet wrote:
> >Here is some work that definately includes bug fixes, fixes for some
> >pathologies I've seen while playing with the new scoreboard code.
> >It's definately not in a ready-to-commit state yet, needs a little
> >cleanup.  But just shout if you see something that looks obscene.
> 
> XXXXOPTIMIZE_TIMEOUTS looks pretty "TV-MA" to me.  !)

TV-MA ?

> 
> >I'm really considering raising SCOREBOARD_MAINTENANCE_INTERVAL to 5s
> >from 1s.  It seems to respond just as nicely to problems.  But timer
> >granularity is now tied to the maintenance interval ... so 5s means
> >timers might be late by up to 10s.
> 
> That seems fine to me, on the shutting-down-children end.  But does that
> mean new processes will spawn only once every 5-10 seconds?  That could
> mean a slow startup.... 

Nope, it starts up StartServers children without any delay between them. 
After a graceful it also starts up StartServers children with only
moderate delay -- it does a 1-for-1 swap: for each dead child from the
last generation it spawns a new child until it reaches StartServers. 

With the exponential spawning and 5s intervals it still seems to respond
fine to spikes ... and in fact is a little less likely to consider a
netscape running 6 network connections as a spike. 

Dean



Mime
View raw message