Can we measure context switches to see how they affect results?

[schainks]

Running the client threads and server threads on the same machine is costing you valuable time during context switches, especially for platforms like OTP. I am not sure all the data is "clean" if running client and server on the same CPU and not keeping them pinned to CPU threads.

Relevant reading:
http://highscalability.com/blog/2013/5/13/the-secret-to-10-million-concurrent-connections-the-kernel-i.html