No subject

when the load average begin to grow, the situation was absolutely
different:
2) all CPUs spend many time in "si"(soft interrupts) - 50%-60% of their
power they spent trying to find needed socket among 80+ thousands of them.
   At the save time the traffic was no so high - 2-3MB/s "in" and
something similar "out".

So the problem in the second situation is not the inability to handle
received packets - which can be solved by adding another network card, but
the HUGE number UDP sockets _in use_ - so many of them that even extended
hash table does not solve it for 100%.
So the question is - why there are so many sockets that are IN USE.
i guess (i cannot be sure cause i don't know internal Asterisk structure)
that many of those sockets should have been closed, but were not closed by
asterisk for some reason.

So i think it might help - please, ask Asterisk people to check the server
and tell they thoughts if so many sockets are normal situation or not. 

Issue History 
Date Modified    Username       Field                    Change               
====================================================================== 
2009-07-31 17:38 falves11       Note Added: 0108484                          
======================================================================