[Mimedefang] Ratio for sendmail and MD processes

David F. Skoll dfs at roaringpenguin.com
Sun Apr 30 20:41:28 EDT 2006


David Nelson wrote:

> I was under the misguided impression that there should be a 1:1 ratio of
> sendmail processes to MD processes.

No, the ration is typically much higher than 1:1.  I don't even bother
capping the number of Sendmail processes.

> I ended up solving the sendmail problem, but I think I created a new
> one.  When I look at md-mx-ctrl, I still see most if not all the MD
> processes sitting idle.

That's fine.

> There's also a sorely uneven distribution of
> the workload across the slaves (status and histo shown side-by-side
> below).

You're misinterpreting the histo output.  Histo shows *how many* slaves
are active during each request, not *which* slaves are active.

Your histo output looks quite healthy.

> If I trim the number of MD processes, does the workload distribution
> become more linear? 

Nope.  MD deliberately does *not* allocate work evenly across slaves.
Instead, it tries to give work to the "oldest" slave.  The reason for
this is we try to keep the same slave busy as possible, to reduce
the likelihood of paging.

> Is there a recommended ratio of sendmail processes to MD processes?  I
> realize "my mileage may vary," but I wonder if I'm even in the ballpark.

10:1 is probably a good starting point.

Regards,

David.



More information about the MIMEDefang mailing list