Re: High-volume server settings

[Steve Hillman <hillman@sfu.ca>]

We have 4 mailbox servers with about 12,000 users on each so far. It wasn't until we did our student migration over Christmas (which took us from 10,000 total users to 42,000) that we had to tweak our thread counts. The HTTP threads weren't even really obvious (and I'm still not 100% sure we were running out), but it *seemed* to make a difference to web snappyness when I doubled them to 500.

The IMAP thread count was more obvious - when we hit the limit, IMAP clients would hang for 10-60 seconds trying to get a connection while a thread became free. Once we doubled the IMAP thread count (we don't have a lot of IMAP users), the problem went away.  But since IMAP connections are held open, it doesn't take many IMAP users to exhaust your threads.

If your servers are sluggish or unresponsive, the problem is probably one of two things - memory or disk. More memory == less disk reads (doesn't help writes much). We run with 32gb on each mailbox server and give MySQL 14gb. Right now, that's theoretically enough for the entire mysql database to fit in RAM

We have a shelf (14 drives) of fibrechannel disks dedicated to each mailbox server for db and index storage. The 'store' space runs fine off of SATA drives


----- "Dmitry S. Makovey" <dmitry@athabascau.ca> wrote:

> On February 17, 2009, Matt Mencel wrote:
> > We have 5 mailbox servers, all have several thousand students/staff
> on them
> > and we're not having problems.  What hardware are you running on?
> >
> > zimbraHttpNumThreads: 250
> > zimbraHttpSSLNumThreads: 50
> > zimbraImapNumThreads: 200
> > zimbraLmtpNumThreads: 20
> > zimbraPop3NumThreads: 100
> > zimbraScheduledTaskNumThreads: 20
> 
> very interesting. from the above I assume you run proxies in front of
> your 
> mailboxes? So you have no problem with settings as low as this? What
> is the 
> primary method of access of you Zimbra? Is it Web Interface or is it 
> POP/IMAP ? I'm curious because ours is mainly Web Interface (out of
> 300 
> people at least 200 using Web Interface). And we had our thread
> numbers 
> bumped. Now I start to think threads are not the issue.
> 
> > Have you checked out this...
> > 
> http://wiki.zimbra.com/index.php?title=Performance_Tuning_Guidelines_for_Large_Deployments
> 
> yes, that's what we were following including bump in number of HTTP
> threads on 
> mailboxes. But apparently out-of-the box is just fine, according you
> your 
> numbers. Thank you for that information.
> 
> -- 
> Dmitry Makovey
> Web Systems Administrator
> Athabasca University
> (780) 675-6245

-- 
Steve Hillman                                IT Architect
hillman@sfu.ca                               IT Infrastructure
778-782-3960                                 Simon Fraser University