Hello,

I’m seeing an unexpected performance drop in Carpet (ET). I ran two simulations 
with identical parameter files (attached) except for:

Carpet::max_refinement_levels = 9     (Run A)
Carpet::max_refinement_levels = 12     (Run B)

During the period I conducted this experiment, all the refinement levels were 
inactive except for the base level. Therefore, I expected to see the same 
performance in both Runs A and B, because the active grid structure is the 
same. However, I realized that Run B is almost 4 times slower than Run A; a 
plot of wall-clock time per simulation time for these two simulations is also 
attached for comparison.

Is this behavior known/expected (e.g., due to bookkeeping, allocation, or 
scheduling costs that scale with the allowed number of levels), or does it 
suggest a configuration issue on my side? Any pointers on what to check, or how 
to profile this within Carpet/Cactus, to understand the overhead would be 
greatly appreciated.

Thank you,
Mahdi Naseri

Attachment: GasCloudC4.par
Description: GasCloudC4.par

Attachment: GasCloudC4_LessRef.par
Description: GasCloudC4_LessRef.par

_______________________________________________
Users mailing list
[email protected]
http://lists.einsteintoolkit.org/mailman/listinfo/users

Reply via email to