At certain seemingly random parts of the day, my CPU usage will spike up from ~10% on 1 m1.large to 70% on 4 large servers. This all happens within 20 minutes and stays this way for a few hours and often while I'm asleep.
Assuming this is legitimate application traffic, how can I track exactly which http requests or cluster of HTTP requests are responsible for the CPU usage so I know what to optimize?
I've had a look at the apache http log but it doesn't help much because the number of requests is basically the same as it usually is.