[Maia-users] lost connection after CONNECT (solved?)

[Blake Hudson]

I performed some benchmarks on amavis (maia actually) and found that the
number of amavis processes does not significantly impact speed when you
are above 2 and below the number where you start running out of RAM and
start hitting the swap file.

As each amavis process takes a good amount of memory (dependent on your
spamassassin rules), I try to run with as few amavis processes as
necessary to leave RAM for other things. I have ran servers with 4
amavis processes for a 1-2k user setup (which usually require 200 smtpd
processes to keep from getting maxed out on a regular basis).

Where more amavis processes might help is when you are accessing RBLs
that might time out or be slow to respond at time (like pyzor, razor,
etc). I have since removed both razor/pyzor from my checks because I
have found them not to be reliable enough for my tastes (can add
anywhere from 2 to 30 seconds per email being scanned).

-Blake



David Sims wrote:
> On Thu, 9 Aug 2007, Greg Woods wrote:
>
>   
>> On Thu, 2007-08-09 at 10:39 -0500, David Sims wrote:
>>     
>>> Hi,
>>>
>>>   Not sure what the recommended relationship between the max number of
>>> postfix smtpd processes is versus the max number of amavisd processes
>>>       
>> [...]
>>
>>     
>>> ... so I upped the number of smtpd processes in postfix from 10 to 50
>>> which seemed to help....
>>>       
>> Unfortunately, this is completely dependent on what hardware you have
>> and how much mail you are processing for how many users, so there is no
>> "typical" answer. At my site, I could build a small supercomputer out of
>> the stack of 64-bit dual-processor hardware RAID systems we have to
>> process e-mail, so I don't have too many worries about capacity. I upped
>> the number of incoming connections to 200 and I run 15 concurrent
>> amavisd processes on each of four scanning nodes. The system is barely
>> breaking a sweat handling the mail for our 1200 on-site users. But if
>> all you have is a single processor Pentium system handling all your
>> incoming mail, you will bring it to its knees with these settings.
>>
>> The best you can do is tweak things a little at a time and see what
>> effect it has, which sounds like what you are already doing.
>>
>> --Greg
>>
>>
>>     
>
> Hi Greg,
>
>   Thanks for the input.... I would have thought that it wouldn't be too
> hard to develop some kind of empirical 'rule of thumb' for hardware vs.
> email volutme vs. number and type of processes that are typically used....
>
>   In the case at hand, I am running a dual postfix maia filter on a 2.4
> ghz P4 single processor with 1 Gb of memory and around 50 users... 50 smtp
> and smtp-amavis processes seem to be doing ok... It's the amavis processes
> that seem to load up the system pretty well.... High process load seems to
> be better than getting the TCP stack in a snit by running too few smtp
> processes.... because of the wait states associated with the TCP
> connections.... i.e., if you have more traffic than available listening
> processes to deal with it, the traffic doesn't seem to serialize too
> well... I am concerned with the idea of having lots of smtp listeners and
> few amavis processes though...  Any thoughts there??
>
>   During an attack it seems like quick rejection of mail destined for
> 'Unknown Users' is very important to maintaining some level of throughput
> of legitimate mail....
>
>   So... Here's a point on the potential chart:
>
> Dual Postfix/Amavis Maia/
> MySQL/2.4ghz/1Gb/50 users ->     50 smtpd processes   5 amavis processes
>
> but I'm thinking that another variable is the average volume of mail
> and wondering what's the best way to measure that?? For network stuff,
> Toby Oetiker's MRTG seems to set the pace. I'm wondering if it might not
> be possible to create some instrumentation that would graph all these
> parameters to make performance tuning more of a cause and effect science
> rather than an empirical process or trial and error..... I wonder if there
> are any SNMP MIBs that would capture this type of info??
>
> Dave
> **********************************************************************
>
> _______________________________________________
> Maia-users mailing list
> Maia-users at renaissoft.com
> http://www.renaissoft.com/mailman/listinfo/maia-users
>