Performance numbers for a DELL R710

Glenn Steen glenn.steen at gmail.com
Wed Jun 3 15:27:50 IST 2009


2009/6/3 Zaeem Arshad <zaeem.arshad at gmail.com>:
>
>
> On Wed, Jun 3, 2009 at 1:58 PM, Martin Hepworth <maxsec at gmail.com> wrote:
>>
>> more children will help - alot of this is suck it and see as
>> performance can vary alot. normal starting point is 5 children per
>> core and 30 messages per batch.
>
> I am currently running 45 processes and around 25 messages per batch. Will
> try to tweak and see.
>>
>>
>> for that sort of level I'd be looking at multiple machines - if you
>> loose this machine or whatever reason what happens!
>
> Getting 2 more for performance and redundancy reasons
>>
>> Also check the RBL's you're running - things like spamhaus will
>> require a licence for thsi sort of amount and too RBLs slows the
>> processing down alot.
>
> I am considering that. Running a local caching server will help limit the
> number of queries made but will definitely look into that. I remember
> reading about Ironport's architecture docs mentioning that it uses an
> asynchronous kernel. Has anyone played with the 2.6 kernel scheduler?
>
Looking to make a 5 time increase in speed (from ~12/s to ~65/s)
you'll likely not only have to look at smart(er) kernel sheduling, but
also consider every aspect of bandwidth saving... In performance
tuning, a job avoided is always better than a job done fast;-).

Cheers
-- 
-- Glenn
email: glenn < dot > steen < at > gmail < dot > com
work: glenn < dot > steen < at > ap1 < dot > se


More information about the MailScanner mailing list