Difference between revisions of "Mohid OpenMP"
From MohidWiki
(→Windows) |
|||
| (One intermediate revision by one other user not shown) | |||
| Line 3: | Line 3: | ||
==How to pre-set the number of threads to use== | ==How to pre-set the number of threads to use== | ||
===Windows=== | ===Windows=== | ||
| − | + | In the Module model it has to be added the following keyword | |
| − | + | ||
| − | where N is the number of threads you want to use. It is recommend that N be equal to, or less than, the number of physical cores of your machine. | + | OPENMP_NUM_THREADS : N |
| + | |||
| + | where N is the number of threads you want to use. It is recommend that N be equal to, or less than, the number of physical cores of your machine. (For example, in a quad core machine, like the intel i7 970, N = 4 gives the best performance.) | ||
The you can launch MOHIDWater | The you can launch MOHIDWater | ||
Latest revision as of 17:26, 28 April 2011
This wiki entry is about the project to parallelize Mohid with OpenMP directives.
Contents
How to pre-set the number of threads to use
Windows
In the Module model it has to be added the following keyword
OPENMP_NUM_THREADS : N
where N is the number of threads you want to use. It is recommend that N be equal to, or less than, the number of physical cores of your machine. (For example, in a quad core machine, like the intel i7 970, N = 4 gives the best performance.)
The you can launch MOHIDWater
exe> MohidWater.exe
Official benchmarking
Benchmarks must be identified with the processor type (intel, amd), the architecture (32bit, 64bit), the number of cores (#2, #4) and the number of threads (#4, #8, ...). The simulation must be given an id such as PCOMS, PCOMS-bio, Azores, etc... The benchmarks must be performed with the data contained in the MOHID Outwatches logs, once with the Model run without openmp, and once with the model run with openmp.
Benchmarking formula
The benchmark formula is:
Speedup = A / B, where A is time with openmp and B is time without openmp. If Speedup < 1, then speed is up; if Speedup > 1, then speed is down.
Official results
PCOMS 20100801 quad-core 4 threads
Module SubRoutine Time w/o Time w Speed-up
ModuleModel RunOneModel 727.679 193.4 0.87
Azores 20100801 six-core 12 threads
Module SubRoutine Time w/o Time w Speed-up
ModuleModel RunOneModel 8792.656 7627.37 0.27
Obs: the Module Life single loop was successfully parallelized.
