[cwl]

Hi everyone.

Got curious - starting Star in parallel on 28 cores (2 x 14, HT off) machine, getting

Starting local server: starccm+ -server -np 28  "D:\Work\star.sim"

************************************************************************************
Warning: CPU contention on 28 processes could not be resolved.
              CPU affinity will be cleared on those processes.

              This is a strong indication that some hosts are over-subscribed with
              STAR-CCM+ processes. Over-subscribing a host is discouraged since it can severely affect performance.
************************************************************************************

MPI Distribution : IBM Platform MPI-9.1.4.4
Host 0 -- NodeA -- Ranks 0-27
Process rank 0 NodeA 5044
Total number of processes : 28

Manual says following:
-rthreads N Client option.
Sets the number of threads N available for visualization tasks, including advanced rendering.
When using this setting, consider the following:
If you specify a high value for render threads, you can incur resource contentions that interfere with the proper functioning of your computer.

Running in parallel on 27 cores (with Number of Client Render Threads = 1 in Options) does not invoke this warning.

Does it atually mean - that I'm supposed to keep some amount (min 1) of cores free?
Or am I missing some specific launch parameters?

[LuckyTran]

Affinity is a broad scheduling issue not specific to Star-CCM.

The warning is that the affinity is not being set when the threads are being launched. This can easily happen when there already is any (other) program running that is consuming a not insignificant cpu-load or not insignificant memory allocation. It's a warning, not necessarily an error. When you launch with less than the maximum number of threads, the threads get assigned to the unused core due to the load-balancing algorithm of the job scheduler and you tend to avoid these warnings. But that doesn't mean that you shouldn't be launching with max thread count.

If you are sure nothing will interrupt Star, you can proceed. But keep in mind whatever process that does have affinity can one day potentially usurp the resources from Star (e.g. your background bitcoin mining code).

[@Arash]

Originally Posted by LuckyTran

Affinity is a broad scheduling issue not specific to Star-CCM.

The warning is that the affinity is not being set when the threads are being launched. This can easily happen when there already is any (other) program running that is consuming a not insignificant cpu-load or not insignificant memory allocation. It's a warning, not necessarily an error. When you launch with less than the maximum number of threads, the threads get assigned to the unused core due to the load-balancing algorithm of the job scheduler and you tend to avoid these warnings. But that doesn't mean that you shouldn't be launching with max thread count.

If you are sure nothing will interrupt Star, you can proceed. But keep in mind whatever process that does have affinity can one day potentially usurp the resources from Star (e.g. your background bitcoin mining code).

Hi Dear friend,

I use an Intel core i9-13900k processor that has 24 physical cores. When I use the number of cores for example 14 or 16 or... in order to parallelize and reduce the solution time, my solution goes slowly. What should I do ?
I selected 16 cores in the Fluent window and then imported the mesh file into Fluent. As soon as Fluent opens in the console, I receive the following message:


Note: Rank = 0: Process affinity not being set (6).
Note: Rank = 1: Process affinity not being set (6).
Note: Rank = 2: Process affinity not being set (6).
Note: Rank = 3: Process affinity not being set (6).
Note: Rank = 4: Process affinity not being set (6).
Note: Rank = 5: Process affinity not being set (6).
Note: Rank = 6: Process affinity not being set (6).
Note: Rank = 7: Process affinity not being set (6).
Note: Rank = 8: Process affinity not being set (6).
Note: Rank = 9: Process affinity not being set (6).
Note: Rank = 10: Process affinity not being set (6).
Note: Rank = 11: Process affinity not being set (6).
Note: Rank = 12: Process affinity not being set (6).
Note: Rank = 13: Process affinity not being set (6).
Note: Rank = 14: Process affinity not being set (6).
Note: Rank = 15: Process affinity not being set (6).


When using the number of cores, for example 14 or 16 or... my solution is slow and very annoying.
Please help me what to do?

Thankyou

[cwl]

Originally Posted by @Arash

I use an Intel core i9-13900k processor that has 24 physical cores. When I use the number of cores for example 14 or 16 or... in order to parallelize and reduce the solution time, my solution goes slowly

i9-13900K has 8 "performance" cores and the rest 16 are "efficient cores".

[@Arash]

Originally Posted by cwl

i9-13900K has 8 "performance" cores and the rest 16 are "efficient cores".

You mean what should I do now?