How to ensure that the grid partition is the same every time it runs？

General
1

Dear all,
I am conducting a strong scalability test on a solver, and with the same code and the same number of rank (128), the output of the solution is inconsistent. It is not that the result is incorrect, but rather the output I specified. What I can think of is the difference in grid partitioning?

2

What is the output you specified? Any chance you could share a minimum example to replicate your issue? The partitioning should be the same across these runs, I don’t think there is a random component.

2 Likes
3

Thanks, the output is solving time of the solver.After my confirmation, it should be due to the lack of exclusive nodes on the supercomputer server that there is a significant difference in running speed under the same code and process.

4

Solving time is impacted by the usage of others on the same system, what processes/nodes you are allocated etc.