The mailing list for listmasters using Sympa

[Olivier Salaun - CRU <address@concealed>]

First of all, I suggest you read the Sympa tutorial that provides some 
optimization information :
http://www.sympa.org/doc.html

elijah wrote:

> As load on sympa increases, what factors limit performance the most? Do
> people have anecdotes, gut feelings, or hard data? Has anyone profiled
> sympa under a stress test situation?
>  
>
> We have reached 70K+ messages a day and realized it is time to build a new
> list machine(s). I would like this new setup to be able to scale to a
> rather large sympa installation. Any advice as to a good strategy would be
> greatly appreciated. I am wondering about the relative importance of ram,
> disk speed, clustering, etc.
We are running Sympa on a single server (bi-Xeon 2.2Ghz / 2 Go ram ).
Our mail traffic is about 7 Million messages / month ; we have 
experienced 500 K messages
a day. As exposed in the tutorial, MTA and DB tuning is essential.

> My superficial look at the problem makes me believe that wwsympa.cgi is
> the bottleneck. It eats up CPU like nobody's business and requires tons of
> ram. My impulse would be to first put wwsympa on its own machine with
> lots of ram and a fast processor and put everything else (including
> sql and disk storage) on another box.
>
> One person in the FAQ entry on clustering sympa suggests offloading
> outgoing SMTP to another machine. This seems pretty pain free, but I am
> unsure as to the benefits. Any experiences here? What about disk storage?
> All that mail must cause sympa to be writing and reading to the disk like
> crazy. Maybe a striped RAID should be priority #1? Who accesses the disk
> more, sympa.pl or wwsympa?
>
> How about this for three machines:
>
> 1) SMTP machine(s): modest disk, ram, cpu. cheapos.
Note that sendmail does lots of file creation/deletion ; disks with fast 
access would be a good idea.

> 2) fileserver/mailserver: run sympa.pl, bounced.pl, archived.pl, postfix
> and mysql. Store all of /home/sympa here. big disks, RAID-5, btree
> filesystem (xfs, reiserfs). invest less in ram and cpu, more in disks.
>  
Note that machine (3) needs to access these disks.
ReiserFS is usefull when having lots of list directory in a single expl/ 
dir.

> 3) webserver: small disk, fast cpu, tons of ram. run apache + fastcgi +
> wwsympa. maybe archives should be on this machine since most people read
> archives from the web.
>
> Good advice might be: "buy the fastest you can." The honest truth is that
> we are a shoestring organization with all volunteer labor and very little
> money. For this reason, I am keen to know where we can get the most
> benefit for the least amount of money. Maybe one really beefy machine
> would be better than three modest ones? We will probably just start by
> replacing our current list machine and migrate to a cluster later, but I
> would like to map out a good path for the future now.
>
> As we do move to a cluster, I will most certainly post to the FAQ a report
> on the results.
>
> Thanks,
> -elijah
>  


--
Olivier Salaun
Comite Reseau des Universites