[Freeswitch-users] concurrent calls capacity planning

[Niranjan kl]

We started using freeswitch for one of our products. We use esl to
control calls. I am trying to come up with numbers to do capacity
planning. I ran my tests on 8 core cpu 16 GB Ram server. This is on
ubuntu 14.04 linux (kernel 3.13) distro on amazon web services. I used
sipp to do load tests. I used freeswitch's external profile (single
profile only) as I will have to deal with third party SIP trunk
providers and G711 is the codec I used. When I executed 500 concurrent
calls by creating 500 calls at  the same time, things work fine. I see
linux's load staying close to 3. When I do 1000 concurrent calls by
creating 1000 calls at the same time, linux's load shoots to 75 and
sometimes reaching 100. Earlier, I did the same test on 4 core 8 GB
server, load shot up to 75 for 500 concurrent calls.
When I searched for freeswitch installation on aws, I read that timer
configuration and low latency kernel changes are recommended. But It
also says it is not needed for 3.x kernel. So, Is this needed to
extract good performance from hardware on aws ? Instead of adding
hardware blindly, I would like to figure out is there any other way I
can profile this and understand the relationship between hardware and
concurrent calls number.

Thanks,
Niranjan KL