Using Trace/Replay on NPB
From the email on the simgrid-user mailing list from Mauricio Palma
Hello,
I am trying to use the "trace replay" feature from SMPI to execute benchmarks from NPB. I am able to create the trace, but I am getting errors when trying to replay some of these traces.
Command to create the trace:
smpirun -np 4 -platform ./platform.xml -hostfile ./hostfile_4 --cfg=smpi/host-speed:3000000000f --cfg=smpi/display-timing:yes -trace-ti --cfg=tracing/filename:myTrace ./cg.C.x
Command to replay the trace:
smpirun -np 4 -platform ./platform.xml -hostfile ./hostfile_4 --cfg=smpi/host-speed:3000000000f --cfg=smpi/display-timing:yes -replay myTrace
Is it the correct way to use the trace/replay feature?
I also noticed that the number of lines of a trace can vary if you recreate it, is it expected? I thought we would always get the same number of lines, just getting variations on the values of "compute". In the particular case of benchmark EP, one of the ranks gets a really different number of lines on its trace (17 lines against ~51K lines for the other ranks).
I am using SimGrid 3.27 and NPB 3.4.
The error is different on each benchmark as reported below.
EP: It works, the timing matches with the run that created the trace, but gives the following error message in the end:
double free or corruption (fasttop) Execution failed with code 134.
IS: Worked but the replay timing does not match to the run that created the trace. (~20 seconds on trace creation and ~33 seconds on replaying it)
CG: following error:
[host2:2:(3) 149.671933] /home/mauricio/Dragons_Lair5/SimGrid_3_27/simgrid-3.27/src/smpi/mpi/smpi_request.cpp:468: [root/CRITICAL] Actor pid=4 is gone?? Backtrace (displayed in actor 2): (backtrace not set -- did you install Boost.Stacktrace?) Execution failed with code 134.
FT: following error:
[host2:2:(3) 7.010946] /home/mauricio/Dragons_Lair5/SimGrid_3_27/simgrid-3.27/src/xbt/exception.cpp:49: [xbt_exception/CRITICAL] Uncaught exception std::bad_alloc: std::bad_alloc [host2:2:(3) 7.010946] /home/mauricio/Dragons_Lair5/SimGrid_3_27/simgrid-3.27/src/xbt/exception.cpp:76: [xbt_exception/CRITICAL] Current backtrace: Execution failed with code 134.
LU: Worked MG: Worked BT: Worked SP: Worked
Thank you,
Mauricio Palma