code path profiler
Thu Oct 21 13:57:07 CEST 2010
Existing tools (eprof, fprof) aggregate by process. This may be useful to analyze some systems but not server apps I usually deal with.
In server apps it's useful to understand where the performance dips are along a path that starts with a request coming in and ends with the response going out. This sets up a timeline and I'd like to know what's happening alongside it.
The timeline includes function calls and messages passed from one process to another. It also includes what the rest of the system is doing during that time (garbage collection?).
Basically, I'd like to aggregate profiling information along a given repeatable code path, rather than per process. Then, I'd like to know the variation between the time taken to server requests during benchmarking.
Does this make sense? Has anyone done anything along these lines?
More information about the erlang-questions