Scaling tests

From Arbeitsgruppe Kuiper
Revision as of 09:25, 22 April 2024 by Dennis (talk | contribs) (Created page with "== Sources == The [https://hpc-wiki.info/hpc/Scaling HPC Wiki] provides a good source for the theory and how to perform scaling tests. == Scaling Tests for Belt == To perform scaling tests for belt, you can use the '''-maxsteps n''' starting option, which tells belt to only run for '''n''' steps and then print out the walltime and time step information. One can then proceed by taking the '''average time/step''' of the different runs and compute the speedup and efficienc...")
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to navigation Jump to search

Sources

The HPC Wiki provides a good source for the theory and how to perform scaling tests.

Scaling Tests for Belt

To perform scaling tests for belt, you can use the -maxsteps n starting option, which tells belt to only run for n steps and then print out the walltime and time step information. One can then proceed by taking the average time/step of the different runs and compute the speedup and efficiency for the strong and weak scaling tests.

Note: There is a bug in the current belt version, which leads to wrong average time/step when restarting a simulation. When restarting, belt measures the new walltime from the restart of the simulation and then divides by the number of time steps since the start of the simulation. To work around this bug, one could change this line in the main.c:

if (g_stepNumber > 0) print ("> Average time/step %10.2e (sec) \n",

difftime(tend,tbeg)/(double)g_stepNumber);