pscedu / distrib.spiobench Goto Github PK

--------------------------------------------------------------------------------
Department of Defense
High Performance Computing Modernization Program Office

TI-06 Benchmarking
Streaming Parallel I/O Benchmark ( SPIOBENCH ) Operational ReadMe

mpt  05/05
--------------------------------------------------------------------------------
Please review the TI-06 Benchmarking Test Description and Rules documentation 
for test descriptions, system configuration, operational, and submission 
guidelines.
--------------------------------------------------------------------------------

Description:
   The Streaming Parallel I/O test measures the maximum I/O performance of the 
   system for streaming applications.

Preface:
   This test must have a shared file system with a minimum of 160 GB of space
   available.

src/spio.c:
   1. If the platform does not support the valloc (page aligned allocation) call 
      then change the call to malloc.

   2. If large file, files > 2 GB, support is not inherent
      the O_LARGEFILE flag may have to be inserted into the open(2)statements.
      No other changes are allowed to open.

Build Instructions:

1. Tar extraction of "spiobench.tar.gz" created the "spiobench" directory.
   "cd" to the "spiobench" directory.

Makefile:

2. To make SPIOBENCH, do the following:

*** for ZEST tests:
   Use the makefile "Makefile.nosys"
*** for "local" execution only:
   Use the makefile "Makefile".

   Review "Makefile" for proper compiler flag settings. 
   Example compile options are included in the makefile.  

   Uncomment the appropriate section of compiler flags, or if none are 
   appropriate for the target machine add an appropriate section in the same
   style as the others.

   Vendors must specify the appropriate libraries during the build process.
   The makefile contains a list of general libraries that may or maynot be used 
   for any individual test.  Please carefully check the build for unresolved 
   external references.

   Make certain the selections used for the final results are appropriate and
   effective.

3. Make sure that the proper flags for compiling in 64-bit mode are present 
   if necessary.

   Type:  "make"
   (If you have GNUmake, 
       modify compiler flags in cmplrflags.mk,
       type: 'gmake all' )

   Use of compiler options and environment variables must be documented and 
   explained in the TI06_benchmark_compiler_flags.xls Excel spreadsheet.
   Selection of the best compiler and linker options is strongly encouraged.

Execution Instructions:

1. Modify the run script:
   Edit the script to create the scratch file directory in a large shared
   file system. (SCRATCH_DIR)

   One script is provided to run the SPIOBENCH benchmark.

      do_spio - runs the spio benchmark

   The test is run at 16, 32, 48, 64 128, 256, 384 and 512 processors.

*** for ZEST tests:
   Make sure that SCRATCH_DIR starts with "/zest/"
   ... other ENV variables have been set accordingly

2. The script is executed from the spiobench directory.

      ./scripts/do_spio
         or
      ./scripts/do_spio 16 32 48 64 128 256 384 512

   The tests should be run on as many processor counts as possible.

3. Runtime for spiobench is determined by your i/o system's bandwidth and 
   performance. An average runtime of 5 to 15 minutes per cpu count is
   normal.

Results:

1. Results can be viewed in the directory "spiobench/results".

2. Tar the entire spiobench directory with the command: "make tar".
   The tar file "spiobench_results.tar" is created in the "synth6" parent 
   directory with all of the results, source code, run scripts, and makefiles
   used to build the tested benchmark.

   Return the "spiobench_results.tar" file as the deliverable for SPIOBENCH.