[CentOS] Random failures with httpd and ejabberd

[Nikita Kipriyanov]

Hello

Our server runs CentOS 5.2, kernel 2.6.18-92.1.18.el5xen x86_64, in the 
domU under Xen, and dom0 runs same version of OS. The CPU is AMD 
Athlon(tm) 64 X2 Dual Core Processor 4400+.

We're experiencing some strange failures with several services, most 
notable web service and jabber daemon.

There is Apache/2.2.3, PHP 5.1.6, and phpBB 2.0.14. It is internal 
forum, we access it via HTTPS, with client certificates.
There is also other PHP application, which stores sensetive data in the 
MySQL database, encrypted with Blowfish.

All is working OK for a while, after that phpBB stops parsing 
[code]...[/code] and [quote]...[/quote] bbCode sections (all other 
bbCode markups are working). At the same time, Blowfish starts returning 
wrong decoding of data.
After the httpd service restart, problems disappear and all works. But, 
after a while it breaks again. Probally, it works while there is single 
user, and breaks just when someone else accesses the server. I can not 
be more precise, at what moment the failure occurs.

Also, ejabberd is working only partially: it randomly loses many 
messages in conferences, and sometimes it returns then 'Message can not 
be delivered, reason: ""' (I can not say the exact error message, as it 
appeared localized).
I also can not set up any message size limits (when they're on, server 
simply doesn't start), and can not setup any transports (to ICQ or 
other). I undertake an attempt to compile it by myself, but nothing was 
changed.

I can not locate any log messages, which are related to this problem. 
So, I need help in diagnosing and solving these problems. Even a little 
ideas of what the problem's root is will help. If I missed some valuable 
information, please, ask for it, I will then provide. I simply don't 
know, for which information to look.

And, please, sorry for my awful english.

Nikita