Hardware Reference
In-Depth Information
amount of time spent in I/O fell from over 26 minutes to 30 seconds, less than
1% of application runtime.
24.6 Conclusion
In summary, in order to get optimal performance from a complex I/O
subsystem, it is useful to do I/O benchmarking and I/O application profil-
ing. Benchmarking allows an HPC analyst to understand the performance
of an I/O subsystem, while I/O profiling shows an application developer de-
tails about an application's I/O patterns that can illuminate how to improve
performance.
Bibliography
[1] CrayPat Performance Tool.
http://www.nersc.gov/users/software/
debugging-and-profiling/craypat/
.
[2] FPMPI-2 Fast Profiling Library for MPI.
http://www.mcs.anl.gov/
research/projects/fpmpi/WWW/
.
[3] HPC-5
Open
Source
Software
Projects:
LANL-Trace.
http://
institute.lanl.gov/data/software/
.
[4] HPC Toolkit.
http://hpctoolkit.org/
.
[5] Integrated
Performance
Monitoring
(IPM).
http://ipm-hpc.
sourceforge.net/
.
[6] Intel VTune Toolset.
http://software.intel.com/en-us/
intel-vtune-amplifier-xe
.
[7] IOR Project Website.
http://sourceforge.net/projects/ior-sio/
.
[8] Lustre File System.
http://wiki.lustre.org/
.
[9] MPI Standard Profiling Interface.
http://www.open-mpi.org/faq/
?category=perftools
.
[10] NERSC Hopper Cray XE6 System.
http://www.nersc.gov/users/
computational-systems/hopper/
.
[11] Valgrind.
http://valgrind.org/
.