A customer using vtune 2013 tried this command line for a 40+ hour long run, 16 MPI ranks - I know, it's asking a lot. Customer reported:
"We've been trying to use it for a problem that runs for 40+ hours/
The elapsed time reported in vtune is far less than it should be, something like 1 hour in this case. Also, the finer-grain times reported when browsing the functions appear to be far smaller than expected.
We found some information by googling vtune, indicating that some options can be set to allow for longer runs and collecting larger sets of data. We tried those without success so far.
If anyone has successfully used vtune with large/long runs, please let us know.
amplxe-cl -r vtune_results -collect hotspots -target-duration-type=long -data-limit=500000 -- mpirun -np 16 <executable> <args>
1. version of Vtune used
Before running, we source this file:
% source /usr/local/intel/parallel_studio_XE_2013/vtune_amplifier_xe_2013/amplxe-vars.sh
We ran on a linux workstation, with 16 intel cores: Xeon E5-2670
3. Serial or MPI:
MPI, 16 ranks
4. Intel's SEP driver or default drivers?
Default drivers I guess, the /sbin/lsmod command didn't return anything about 'sep'.
5. Type of collection:
6. How long after the run did we wait:
Quite a while, it looked like it completed successfully.
"The elapsed time reported in vtune is far less than it should be..." Yeah, that is because of the data limit. Once the data limit is hit, VTune Amplifier will stop collecting data. The Elapsed Time displayed is only for the data collected.
I would start with the old adage, "Don't do that!" There is no way you are going to collect 40 hours worth of data and be able to analyze it. We have a data limit set by default for this very reason. What is the problem they are trying to analyze? Why can't they collect data for a few minutes during that 40 hours and analyze what is happening? I highly recommend you help them figure out what they are trying to analyze and narrow the collect.
Since you are using "hotspots", the drivers don't come into play. It is all software based - no hardware drivers.
Which Linux and why don't they upgrade to the latest release?