Increase Performance / Separate Horde from Mail Server on Server with 12 000 Accounts

[eugenevdm]

We have major performance problems on our mail server with 12 000 accounts.

We wanted to go for a distributed Atomic or 4PSA solution but both are not compatible with Plesk Expand.

In order to isolate the problem we want to try and separate Horde from the mail server. In essence create a front-end server for Horde that connects to the back-end server.

Alternatively we need some advice on splitting IMAP/POP/Port 25/etc.

The only thing we're doing is using an SMTP forwarder which seems to alleviate some load, but still we have problems with performance.

We've given up for two years to use Spam Assassin, this completely kills the server.

Can anyone advise us? I'll pay for this service.

 

[burnleyvic]

Here's a free hint, although I don't know if this is what you're looking for: check /etc/psa-webmail/horde/imp/servers.php
You should perhaps try to give more information on what "major performance problems" actually means.

 

[eugenevdm]

Thanks for the reply. The main issue is the server crawls to a halt.

* SMTP timeouts
* Apache (Horde) dead slow

1000s of processes such as:

30 482 1 0 12:53 ? 00:00:00 /usr/lib/plesk-9.0/postfix-queue 127.0.0.1 10027 before-queue
30 484 1 0 12:53 ? 00:00:00 /usr/lib/plesk-9.0/postfix-queue 127.0.0.1 10026 before-remote
30 485 1 0 14:38 ? 00:00:00 /usr/lib/plesk-9.0/postfix-queue 127.0.0.1 10027 before-queue
30 487 1 0 14:38 ? 00:00:00 /usr/lib/plesk-9.0/postfix-queue 127.0.0.1 10026 before-remote

What I'm after is architectural advice on how to split services, this would be ideal:

* Server 1: SMTP incoming with inbound SPAM protection
* Server 2: POP3/IMAP (separate server)
* Server 3: Webmail (separate server)
* Server 4: SMTP outgoing relay host (this is easy and already implemented)

check /etc/psa-webmail/horde/imp/servers.php

This is good advice because theoretically I can modify those files to retrieve Horde mail from another server.

 

[eugenevdm]

While I'm loosing clients might I ask this question. Based on the "official" Postfix stress document here:
http://www.postfix.org/STRESS_README.html

It says:

Reduce smtpd_timeout (default: 300s).

On our Plesk Postfix server we have:

smtpd_timeout = 3600s
smtpd_proxy_timeout = 3600s

Might Parallels have changed the default from 300 to 3600 (1 hour!)??

That means every connection to the server can keep it open for one hour??

 

[OlegN]

Timeout has been increased because some users experienced problems with receiving emails of big size (more than 300Mb)

Since Plesk 10.4.4 performance of mail subsystem was significantly improved after changing scheme of mail delivering in Postfix from using Plesk mail proxy to Postfix milter (mail filter, http://www.postfix.org/MILTER_README.html)

 

[eugenevdm]

To make a long story short we have to reboot our server about once a month for this problem to go away. I guess there some kind of memory leak or something.