Amneet Bhalla
Amneet Bhalla
I am now running a test in which I reinitialize the Stokes solver after every 5 time steps instead of re-initializing it at every time step. This test can check...
@drwells: I like the epic label for this issue.
@knepley : Are you aware of PETSc-based large scale apps in which folks tear down and build PETSc solvers at every time step instead of updating matrix coefficients? Have they...
Ok, so the simulation is around ~20K time steps. It has slowed down, but not entirely clogged yet as I used to observe before. So it appears that PETSc repetitive...
> I have not noticed any slow-down on any of the simulations/applications that I've been running recently. They all run smoothly for the entire 7 days of the simulation and...
That's good the regular INS did not show any slow down. It indicates that SAMRAI vectors used in PETSc solvers are not responsible. I have tested my app on Comet...
> @amneetb The solver itself is a tiny amount of memory, so if something is going on it must be about work vectors or a matrix. Can you make the...
@Kaustubh-Khedkar is running the same app on my start-up cluster using 4 nodes each with 32 CPUs and 64 GB RAM. The solver is reinitialized at every time step. It...
Hmm ok. For 128 cores we are not observing any slowdown as well. Unfortunately I can't run the app with 256 cores on my cluster (yet) to test it on...
Yep, the compute time is increasing. Atleast it is not just the multiphase solver that shows slow down. Houston we have a problem ;-)