Processes clump together on one CPU while the other is left idle

[Fallaxia]

Dear Forum,

On my Win 11 Enterprise machine (ver 27924) I use Process Lasso to distribute BOINC EInstein@Home Tasks evenly across both CPUs. Each CPU has 64 Cores and 50% shall be utilized, in other words 64 Instances.

If I leave them untouched, they clump all together on either CPU 1 or CPU 0 after a few hours.
If I use Instance balancer I can get them spread across both CPUs, but other issues arise like performance drops.

What I am looking for is a simple way to assign an instance to a CPU socket or NUMA node, not to micro manage them on a per core basis.

Once BOINC is started PL will assign the 64 instances and put 32 on CPU0 and 32 on CPU1, easy.

Is this function already available or any plans to get it included any time soon?

[Jeremy Collake]

I'd be curious to hear the results of running the Group Extender on your BOINC instances. See menu item "Options / CPU / Group Extender". Let me know how it goes!

[lilysmith1098]

You can already do that in Process Lasso by setting the CPU affinity to whole NUMA nodes or sockets instead of individual cores, which should keep your BOINC tasks balanced without manual micromanagement.

[Fallaxia]

I'd be curious to hear the results of running the Group Extender on your BOINC instances. See menu item "Options / CPU / Group Extender". Let me know how it goes!

I am currently testing the Group Extender Feature, will let you know the results shortly.


@lilysmith1098

Please explain how this is achieved. Lets say 4 MilkyWay@Home Instances are running, each requiring 16 CPU cores. How do I manage to get them spread over the two NUMA nodes / CPU Sockets evenly in the CPU affinity menu? While I can select individual CCDs I don´t see how this will avoid clumping them together ince all the instances are affected by the CCD choice.