Hi Guys,

I have a basic question here.
If a process is being executed by a processor core and the cpu utilization of that core goes really high, does the process fall back to another core which has less utilization. The process is not a a multi-threaded process. And the OS is RHEL 6 running on a VM.

Regards,
Joemon

Optimally, one process, which uses the computing power of one core completely, does not change the core.

I think, this is done by the kernel load balancing routines.

No. Would cause a cache flush for no gain.

All bets are off running under a hipervisor.

Windows switches a single threaded process frequently among idle cores, despite the performance loss from cache effect. I think Linux does also, but I'm less sure. I think the motivation is thermal balance.

When I run a high-load single-thread task I can actually see it jumping from core to core on my machine, so I am not quite sure if your statement is correct. Can you elaborate on that?

Welllll .... that's the theory ...

Several years ago I attended a presentation where the speakers had patched the Linux scheduler to better balance core usage. They had analysed process migration across cores/nodes, and reckoned they could do it better.
I waited for the patchset to be released as promised but I never found it. All this was just prior to CFS being merged upstream.

My observations are that the scheduler seems to do a pretty good job, but I admit to not having benchmarked it. Pre-emptable kernel threads and containers have made things a little more obscure too. I have a to-do list that includes some perf analysis of task movement across cores.