Skip to content
Snippets Groups Projects
Commit be00b851 authored by Gab's avatar Gab
Browse files

added test cases README

parent ac701032
No related branches found
No related tags found
No related merge requests found
Test case presentation
======================
This case is defined as the RUN_NB640.
It is small run to validate the execution.
This test is used to evalutate the performance of the computer. May scale up to 80000-100000 cores or up to 640GPUs. 32x32x10=10240 domains are defined.
Reference time for 10 days simulation on Jean-zay :
* CPU No I/O : 3000s on 5120 cores (128 nodes)
* GPU No I/O : 782s on 160 GPUs (40 nodes)
Test case presentation
======================
This case is defined as the RUN_NB40.
It is small run to validate the execution.
Reference time on Jean-zay :
* CPU No I/O : 38s on 800 cores
* GPU No I/O : 80s on 4 GPUs
Test case presentation
======================
This case is defined as the RUN_NB40_XIOS.
It is small run to validate the execution. Comparison can be done with or without IO servers to look at the impact on the performance.
Reference time on Jean-zay :
* No I/O : 38s on 800 cores
* With IO server : 57s with 800 cores for dynamico and 8 xios servers
Running with XIOS IO servers requiered to execute run in MPMD mode, i.e. with dynamico and xios io server independant executable.
Typically the running command for OpenMPI of type :
mpirun -np 40 icosa_gcm.exe : -np 20 xios_server.exe
IO server processes must be spread all along the node reservation. We advice one IO server by node, shared with icosa_gcm.exe executable.
XIOS is not multithreaded, so the difficulty is to merge on the same node the multithreaded dynamico model. Be carefull about the binding and placement.
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment