SCHEDULE and high load situations

[Jeff Roberson]

On Wed, 11 Aug 2004, Martin Blapp wrote:

>
> Hi,
>
> >
> > You might well want to try 4BSD.
> >
>
> I did that too. The milter stress test I run was sending 200 mails
> with 5 different sorts of attachements into a mail loop. This means
> these 200 mails are going 26 times trough the milter.
>
> The ULE scheduler did process them first very fast. With more processes,
> the sendmail transactions lagged a lot and it was only running 1-2 of
> them at one time. This sucks because there is also a lot of timeout
> handling (waiting for DNS responses).
>
> All in one I must say that the SCHED4BSD processed the mails in half of
> the time as SCHEDULE did. The mix involved included everything:
>
> - Preforked mimedefang workers
> - Forked Sendmails
> - Threaded applications like clamd, mimedefang-milter
>
> So I'd call it a typical real world situation.
>
> Another thing I observed was that SCHED4BSD has zero IDLE time, while
> SCHEDULE always idled between 20% and 50% ! This with 500 running processes
> and a load of 2.5 -3.0.
>

Hi martin.  This is a great test.  Can you try it with my most recent
change to sched_ule.c?  Your version should be 1.21.  I found some serious
issues that have been addressed.

Secondly, what kind of machine is this?  Is this test simple to setup so
that I may reproduce it here?

Thanks!
Jeff

> Martin
>
> _______________________________________________
> freebsd-current at freebsd.org mailing list
> http://lists.freebsd.org/mailman/listinfo/freebsd-current
> To unsubscribe, send any mail to "freebsd-current-unsubscribe at freebsd.org"
>