httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Chuck Murcko <ch...@telebase.com>
Subject Apache slowness (was Re: fairy nuff)
Date Thu, 30 Nov 1995 17:55:45 GMT
Randy Terbush liltingly intones:
> 
> > Chuck Murcko liltingly intones:
> > > 
> > > It seems like the BSDI sluggishness needs to be addressed, for one thing.
> > > We've started seeing it here, and I know my syslogd isn't backed up its
> > > queues. For the record, it seems like 100k hits/day can induce it.
> > > 
> > Some more data. I run a stock build on the machine in question, except
> > for -DMINIMAL_DNS. The problem seems to occur only when I have at least
> > 60-70 children spawned, but not necessarily servicing connections. I'm
> > chartered to dig into this here until it's fixed, and I haven't ruled out
> > BSDI 2.0.1 yet, either. Hopefully, more later.
> > 
> 
> Interesting. I think the listen queue is hardcoded at 64 for BSDI 2.0.x
> 
Actually, SOMAXCONN_DFLT is 32, but that's been raised to 64 for the kernel
on this machine. We're not getting connects refused, however. The thing is
that it takes us about 14 hours to see the problem at our hit rate, and
Jim sees it about 7 times sooner at 7 times the hit rate. Also, Apache
seems to go deaf at non-maximum traffic times. Here, in the time up to
seeing the deafness mode, we routinely run fine through periods of 60 or
so ACTIVE (handling connects) servers, but we die with no more than 10
or 20 active.

As an initial workaround for Jeremie, he may want to run Apache with 
MaxRequestsPerChild set to a nonzero value, unless he has already tried
running it that way. 100-200 should do the trick. That seems to alleviate
the problem here. I'll know a bit later today for sure.

chuck
Chuck Murcko	Telebase Systems, Inc.	Wayne PA	chuck@telebase.com
And now, on a lighter note:
"The bland leadeth the bland and they both shall fall into the kitsch."

Mime
View raw message