Hello,
 
After days of debugging, I'll need some help from the SER specialists to troubleshoot this problem.
The SER version in service is: 0.9.9+cvs20090925
It's running on a Red Hat EL 5.3 server.
 
I currently have about 10.000 users registered.
I encounter several seconds (1 to 5) without replies from the SIP proxy every 30 seconds.
Some requests are delayed, but most of them are not replied at all.
At the same time, all the 6 SER childs (forking is enabled) get about 100% used CPU (from "top")
 
After some troubleshooting, I found that the rythm of those deny of service was exactly the time configured as the "timer_interval" parameter of  "usrloc" module.
When changing the parameter, problem rythm changes as well.
 
What we did so far:
- Check DNS and MySQL performances
- Moved the number of childs from 6 to 9
- Changed the "timer_interval" ("usrloc" module) from 60 to 900
- Stopped the "natping" feature ("nathelper" module) as it impacted the service as well.
 
Please let me know if the problem has already been observed and what would be the best way to go deeper in the troubleshooting.
I need to solve that as soon as possible: some of our users higly suffer of random REGISTER not beeing taken into account.
 
Sincerely,
Igor