[Spce-user] SEMS SBC CPU question

Andrew Pogrebennyk apogrebennyk at sipwise.com
Fri Feb 14 04:52:05 EST 2014


Hi Larry,

We've seen that 10 media processor threads create some CPU load even
when sems is idle on the VM, so for development VMs we decreased it to
1. You could also try changing the session_processor_threads.

Other parameter is sip_server_threads, but AFAIK it's just a number of
SIP UDP receiver threads and it's 8 by default which is quite
reasonable. Please try to watch the size of Recv-Q and Send-Q in netstat
-anu as you run the test. You'll see if the bottleneck is in the number
of SIP or RTP receiving threads.

P.S. Here's the description from sems core sems.conf.sample:
############################################################
# tuning

# optional parameter: session_processor_threads=<num_value>
#
# - controls how many threads should be created that
#   process the application logic and in-dialog signaling.
#   This is only available if compiled with threadpool support!
#   (set USE_THREADPOOL in Makefile.defs)
#   Defaults to 10
#
# session_processor_threads=50

# optional parameter: media_processor_threads=<num_value>
#
# - controls how many threads should be created that
#   process media - on single-processor systems set this
#   parameter to 1 (default), on MP systems to a higher
#   value
#
# media_processor_threads=1

On 13/02/14 18:06, Schaefer, Larry wrote:
> Andreas,
> 
> Thanks for the response... I have been modifying media_processor_threads in sems.conf, but so far it does not seem to make a difference.  I am still wondering if running in Amazon EC2 is part of the issue.  
> 
> Wondering if anyone has experience with running SEMS in EC2 and if they are seeing any similar problems with CPU.  
> 
> Any additional advice or suggestions would be appreciated...
> 
> Thanks,
> Larry 





More information about the Spce-user mailing list