Wietse Venema:
> Victor Duchovni:
> > On Wed, Dec 15, 2010 at 01:37:48PM -0500, Dave Brodin wrote:
> >
> > > I ran the following command:
> > >
> > > time /usr/local/bin/smtp-source -s 10 -l 10120 -m 500 -c \
> > > -f t...@bluemarble.net -t dbro...@bluemarble.net localhost:25
> >
> > OK, this is smtp-source with 10 (modest) parallel sessions,
> > 10KB (modest) messages, total 500 messages. This should typically
> > accept mail at 100+ messages a second finishing in under 5 seconds.
> >
> > > And got the following output at the end:
> > >
> > > real 0m58.261s
> > > user 0m0.055s
> > > sys 0m0.126s
>
> For what it's worth, here is one data point for Postfix 2.8-20101210
> on FreeBSD 8.2-Beta1 i386 with 2 CPUs running as a VirtualBox guest,
> with the smtp-source program in the VirtualBox guest. This delivers
> mail to /dev/null, and ps(1) shows 10 smtpd processes. The top(1)
> output spikes briefly.
>
> /usr/bin/time ./smtp-source -t dev-n...@localhost -s 10 -l 10120 -m 500 -c
> 168.100.189.20:25
> 500
> 3.29 real 0.00 user 0.11 sys
>
> I don't have a lot of hardware lying around for bare metal testing
> but I can do a temporary install on a 64-bit 2-CPU laptop computer.
Second data point: the test takes a similar amount of time on a
bare metal machine (Postfix 2.8-20101210, FreeBSD 8.2-Beta1 amd64,
Core 2 Duo CPU). Both Postfix and FreeBSD were installed "out of
the box" without any additional kernel tuning or Postfix tuning.
I may be able to do some quad-core tests in a week or so, but I
would be surprised if the result will radically change.
Wietse
