On Fri, Aug 12, 2011 at 1:14 PM, Jain, Deepak (GE Healthcare)
<DeepakJain@xxxxxx> wrote:
Try to use valgrind’s callgrind tool for investigating where the time has been use.
The 75 seconds were lost between network (tcpdump shows what packet with GET query was successfully sent from nginx to http server) and http server callback function, which was called by libevent after 75 seconds (http server log file indicate about this). So, 75 seconds where lost somewhere in libevent.
Callgrind and other profilers shows total time of each function execution, they does not know how long one specific request was in event loop of libevent.
I think callgring will not help.