Application: Gyro
Platform: Teragrid
Tool:PerfSuite
Teragrid Platform information:Info
(Obtained using psinv - PerfSuite machine information utility)
Default mode: Counting (Profiling also available); Default Configuration file is used.
Note: PerfSuite's psrun can be used to perform aggregate data collection ("counting mode") or to do statistical profiling ("profiling
mode"). By default, psrun operates in counting mode.
Usage: The application's batch job script needs to be modified by including the full path of "psrun" command before the application executable when giving these as parameters to mpirun. Eg. Original application: mpirun -machinefile $PBS_NODEFILE -np 16 $GYRO_DIR/src/BigScience
To use PerfSuite: mpirun -machinefile $PBS_NODEFILE -np 16 /usr/projects/perfsuite/bin/psrun $GYRO_DIR/src/BigScience
Note: Each processor emits it's own xml file. The brief summary is generated by psprocess -b BigScience*.xml. They were combined using psprocess -c *.xml. The html results for a single process are generated by 'psprocess -H process_xml_file'.
I'm trying to get the combined results for all processes in HTML. I'm currently blocked on an error on that aspect and communicated this to the PerfSuite developers. Update: Got this message from the PerfSuite team: "I'm afraid that at this point, HTML output is supported only for single-run <hwpcreport> documents, gathered in counting mode."
PerfSuite Output:
B1-std: Brief Summary| Single process(html) | Combined results(XML)
B2-cy: Brief Summary | Single process(html) | Aggregated results(XML)
B3-gtc: Brief Summary | Single process(html) | Aggregated results(XML)
Graphical tools: Currently Vprof,another tool can be used to visualize profiling data generated by PerfSuite. Another approach is to generate comma-separated values(using psprocess --csv) and use a spreadsheet program like Excel.