Hi * lovers,
We have done some perfomance tests with sipp.sf.net.
The testbed has been a mix between AMD Opteron and Intel Xeon with one or two CPU’s and with CentOS X86_64 OS.
We haven’t noticed any difference between one or two CPU’s and my question is:
Is there a compiler parameter to set so asterisk utilizes the second CPU or doesn’t Asterisk benifits more than one CPU ? Couldn’t find any leads on Google/Voip-info.
/Mats
I would doubt that asterisk is purposly only using a single CPU. My best guess is that some other resource is causing a bottleneck. You should be able to easily check to see if RAM, NIC bandwith, or Harddrive IO spikes are causing this bottleneck. Harder things to determine are whether or not the motherboard BUS is a limiting factor, and others like that. If you can post some stats on what you are seeing including test hardware specs, CPU loads, RAM usage, bandwith usage, and disk queue lengths it might be helpful to others.
Scott
Thx for the advice.
I disabled :
set verbosity 0
sip no debug
And changed logger.conf to log less
and added NoCDR to extension.conf
and changed the ethernet to a 1Gbps switch from 100Mbps.
The changes improved the call rate with 20% on the AMD machine but not on the Intel XEON.
And the hardware that has been tested is HP DL140/145 and Sun X4100 servers.
And the performace has been meassures with “vmstat” and ongoing calls te hear the quality. and the sound quality has degraded when vmstats show that the process queue has increased on the CPU’s.
But I have failed to validate that it uses both CPU’s.
The tools top, vmstat, mpstat isn’t giving me numbers that validate that Asterisk utilize both CPU’s.
Advices are welcomed.
/Mats