monitor CPU load on server

[aglyzov]

In addition to monitoring memory load we need to monitor CPU load.

[jlouis]

That is indeed a nice idea. While at it, I suggest we monitor network throughput as well. The reason is that all of these 3 parameters, memory, CPU or Bandwidth will hurt a lot if they are limiting factors.

For instance, if you can saturate the network and run "as many conns as possible", then you are really measuring the size of headers in different systems. This means that we end up with the server which does the correct full headers being the loser of the test.

[ericmoritz]

Ok, I'll add those, it's pretty easy to add to server_manager.py

[ericmoritz]

Hey, not doing the correct thing does wonders for MongoDB benchmarks :p

[ericmoritz]

Do you have any idea how to measure bandwidth saturation?

[jlouis]

Only on FreeBSD :/

[aglyzov]

How about this: http://www.frenchfries.net/paul/tcpstat/

(available in ubuntu)

[ericmoritz]

Here's a dump of netstat -s after a number of connection timeouts for the Java webbit server. Is there anything glaring that anyone sees?

https://gist.github.com/3019940