Nevermind that -t comment, i just looked at the manpage and the usage of -R
removes the potential for that having anything to do with it.

----- Original Message ----- 
From: "ross mueller" <[EMAIL PROTECTED]>
To: "Matt Sergeant" <[EMAIL PROTECTED]>
Cc: <[EMAIL PROTECTED]>
Sent: Wednesday, November 05, 2003 5:20 AM
Subject: Re: High volume problem


> Yeah, i've toggled around with the tcpserver options, even going as far as
> turning off TCPREMOTEHOST and TCPREMOTEINFO lookups without any promising
> results.
>
> Here's the exact runfile i'm using
>
> #!/bin/sh
>
> QMAILDUID=`id -u qmaild`
> NOFILESGID=`id -g qmaild`
>
> #exec /usr/local/bin/softlimit -m 25000000 \
> /usr/local/bin/tcpserver -c 150 -v -p -R -t 20 \
>   -u $QMAILDUID -g $NOFILESGID `head -1 config/IP` smtp \
>   ./qpsmtpd 2>&1
>
> What i find interesting about the issue is even if i set -t to 3 seconds,
is
> that it ends up still having the delay.
> Shouldn't it drop right in if after 3 seconds it can't resolve the remote
> host?
>
> We're currently running a bind 9 caching nameserver locally on one machine
> and dnscache on another to test thier results and neither seem to be
getting
> hit too hard at all.
>
> I'll look into selectserver and see how it runs in comparrison.
>
> ----- Original Message ----- 
> From: "Matt Sergeant" <[EMAIL PROTECTED]>
> To: "ross mueller" <[EMAIL PROTECTED]>
> Cc: <[EMAIL PROTECTED]>
> Sent: Wednesday, November 05, 2003 2:50 AM
> Subject: Re: High volume problem
>
>
> > On 5 Nov 2003, at 3:18, ross mueller wrote:
> >
> > > Anyone have any ideas what may be causing this delay?
> >
> > Hard to tell without full tcpserver command line. Make sure you're not
> > doing slow things like identd lookups (you say you have -R which turns
> > that off, so it's probably something else).
> >
> > If it's not that, do you have a local caching dns server (I use djb's
> > dnscache)?
> >
> > You might also want to consider the new SelectServer instead of
> > TcpServer (in CVS) which handles many more concurrent connections
> > (though will probably break at around 1000 and you'll need to up your
> > ulimit) and probably performs slightly better. Though it's slightly
> > premature at the moment - might need either some extra hands or another
> > couple of weeks before it's absolutely solid, but I've been running it
> > happily here now for a few days (processed a few thousand mails through
> > it).
> >
> > Matt.
> >
> >
>

Reply via email to