OSEC

Neohapsis is currently accepting applications for employment. For more information, please visit our website www.neohapsis.com or email hr@neohapsis.com
 
Subject: Re: high capacity configuration nightmares
From: Pere Camps (pereulivatar.com)
Date: Wed Oct 25 2000 - 13:14:38 CDT

Wietse,

> If you have shells that can't fork, that is a Linux kernel problem,
> and you have to fix that before you can measure Postfix performance.

        That problem has now been solved. The Linux machine is not running
out of processes now.

> To control the inflow, reduce inbound parallelism in the client,
> or reduce the number of Postfix SMTP server processes in master.cf.

        I've triend with 100 SMTP server processes and it didn't work. I
guess it's a disk problem then.

        I'll try with the default 50 SMTP server processes to see if in
that way the disk || cpu is not that much clogged. I'll be running
performance monitoring tools in order to see finally if it's a cpu || disk
thing. At least it is not a memory thing.

        If that doesn't help, I reckon that less than 50 SMTP server is
not going to help. What do you think?

        Unfortunatedly I can not reduce the paralelism of the clients.

        If it the 50 SMTP server fails, I can think of:

        a) The disk || cpu. If the performance tools say that's the
problem, then I'll need to get either more boxes, bigger boxes or a
different configured (hardware wise) box.
        b) Linux. So far it has not given me this kinds of problems, but
it looks like I'm running into something that the Linux kernel is not able
to cope with. Any insight from any members of the list? Any previous
experiences?

        Anyway, thanks for your help,

-- p.