httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Gerald D. Anderson" <gan...@voyager.netcomi.com>
Subject Re: To starve or not to starve (Starving sockets bug) (fwd)
Date Tue, 22 Jul 1997 08:28:01 GMT
Thanks for the responses guys, I did go ahead and submit a PR.  Well, good 
question Marc.  We have kind of an interesting setup.

We actually have a listen directive for each virtual.  Reason being is we run 
8 copies of httpd on each machine with a unique config file for each. This was 
our way of escaping file discriptor problems, there are 300 to 700 virtuals
on each host.

As far as how busy. . .here's a snapshot from 'top' on one of our machines

  2:26pm  up 17 days, 12:13,  1 user,  load average: 0.69, 0.75, 0.60
142 processes: 138 sleeping, 1 running, 3 zombie, 0 stopped
CPU states:  4.2% user, 49.3% system, 48.8% nice, 47.1% idle
Mem:  128096K av, 125748K used,   2348K free, 100884K shrd,  65292K buff
Swap: 130748K av,      8K used, 130740K free                 25008K cached


Not terrible.  BTW this machine I JUST encountered the problem on.  A HUP and
everything's OK.

Thanks again,

Gerald
> In addition to what Dean said, how many Listen directives are you using?
> Most servers should be able to get by without any.
> 
> > ---------- Forwarded message ----------
> > Date: Tue, 22 Jul 1997 11:16:47 +0600
> > From: "Gerald D. Anderson" <gander@voyager.netcomi.com>
> > To: apache-bugs@apache.org
> > Subject: To starve or not to starve (Starving sockets bug)
> > 
> > Hey folks,
> > 
> >   Had kind of a quick question.  We are currently running apache on PPro 200s 
> > runing Linux 2.0.(29,30).  We have between 300-700 virtual sites per machine 
> > (34 machines).  Problem is that we are seeing a problem very similar to the 
> > starving sockets issue (PR#467), however when I examine the src/conf.h the 
> > serialized accept parm. is defined by default for Linux.  My question is, that 
> > if serialized accept is defined is that a DEFINATE that we are not seeing the 
> > exact same problem( which sounds the case to me), or is there some way we 
> > could still be seeing the same starving sockets issue.  The symptom is servers 
> > seeming to randomly quit servicing a port.
> > 
> > If I can be of any other assistance, please let me know.  Otherwise thanks!
> > 
> > Gerald Anderson
> > Sys Admin.
> > Netcom Interactive
> > (972)481-2972
> >  
> > 
> > 
> 

Mime
View raw message