I have been attempting to profile a large-scale application my company builds in order to try and find out what some of the bottlenecks in performance are. To do this, I added -pg to all the compile and link time options in the app. This seems to work some of the time, in that when the app runs to completion and terminates, the gmon.out files have been quite useful (I have other issues with these I won't go into in this thread). The app is generally built on SLES 11 SP1 to run on SLES and RHEL6 systems. The first problem we saw was that on RHEL6, the app will finish its run but never terminates. We know this because the informational messages it puts on the console indicate that it is done and wrapping up execution, but then the app hangs for hours. In one instance it was still hung 24 hours after it announced it was done and we eventually killed it. We tried using a SLES system next, and for one set of test runs it worked fine. Subsequent attempts to run the profiling version on SLES have failed in the same manner as those on RHEL6 - long hang after execution is done. With a small dataset, one such run actually completed - one hour after it said it was done. (The delay between announced completion and return to the command line is usually a few seconds.) We do not have gdb on these systems at all, though I can press to have it installed if that will help. Is there any history of this sort of problem with which anyone here is familiar? I checked the archives and saw only one hit for profiling, and that was not the same (though somewhat helpful). The app is an ~800,000 line C++ program for backing up and restoring data. Any assistance would be most appreciated. Thanks. MR
