Skip to content

clement-parisot/g5k_bench_flops

 
 

Repository files navigation

g5k_bench_flops

automate flops benchmarking of grid5000 clusters

installation:

running

In brief:

  • adapt packages dict if needed (if using different versions of atlas, hpl, openmpi) in common.py

  • run g5k_prepare_bench_flops to precompile each packages on all clusters (the precompiled packages will be downloaded in directory preparation/)

  • run g5k_bench_flops

  • wait...

Details:

  • Run g5k_bench_flops and g5k_prepare_bench_flops with --help to see list of options

Example:

For example. Let's bench the newly installed cluster paranoia in Rennes, not yet in production (as of April 3, 2014):

On a grid5000 frontend or node:

$ ./g5k_prepare_bench_flops paranoia -ML -o "-q testing" -T

$ ./g5k_bench_flops paranoia -ML -o "-q testing" -T

analyzing

When all cases have been run, use the following script to obtain results:

$ ./parse_bench_results <run_dir>

and fill the Reference API.

About

automate flops benchmarking of grid5000 clusters

Resources

Stars

Watchers

Forks

Packages

No packages published

Languages

  • Python 77.7%
  • Shell 22.3%