Desperately trying to debug poor spam scanning performance

[Ben Tisdall]

Steve Freegard wrote:

> 
> IMHO - Pyzor isn't usable anymore unless you are low-volume and can put
> up with the timeouts decreasing your scanner throughput, so I always
> either don't install it or disable it.
> 
> Even though it appears to be unwell on your machines anyway - disabling
> it will prevent the need for the code to get loaded anyway.


Interesting to hear your take on this, anyone else share this view?

> 
> 2)  Compare like-for-like.
> 
> Currently - each machine has different version of Perl modules (some
> newer some older).  Check for updates to each of them on the new machine
> and get the latest versions.

Sure, but I don't think the numbers I'm seeing can be explained away by
module version differences.

> Also - you're looking at the 'Log Speed' output on both machines and
> because they are showing different lower numbers you're jumping to the
> conclusion that something is wrong.... it could be - but the only way to
> be certain is to process the *same* batch of messages on both machines
> (without any other traffic running at the same time) and then comparing
> the results.
> 
> I'd expect the bytes throughput shown in the logs to vary greatly for
> each batch due to the fact that some messages are larger and more
> complex than others, so unless you are running the same batches through
> - then you can't really know for sure that one is slower than the other.

That's exactly what I'm doing - relaying a message throught the test box
to my home box & comparing the figures. I correlate the ms logs with the
exim logs to make sure I'm comparing correctly.
> 
> 3)  Do you have anything configured in 'Spam Lists' or 'Spam Domain
> List' on either machine in MailScanner.conf??

Tried turning off, still sucky.

> 
> 4)  Have you mounted /var/spool/MailScanner/incoming on tmpfs?

On this box not really an option, can't upgrade the RAM beyond 2G & it
has to perform duties other than MS :(

I'm strongly tending towards the theory that I/O is crappy on this box.
I read something not very complimentary about the smart array 5/i on
Linux & certainly the bonnie++ results are worse than those for my home
box (2 x 15K SCSI on the test box, 2 x 7.2K SATA at home).

In all likelihood I'll now be given a new box for MS with enough RAM to
do incoming on tmpfs :)

Thanks for your suggestions Steve.

Best regards,

Ben.

-- 
Ben Tisdall
Linux Systems Administrator | www.photobox.com