On Thu, Feb 02, 2023 at 10:41:03AM -0000, Stuart Henderson wrote:
> While the main reason for OpenBSD not scheduling work on 'cores' it sees
> as SMT rather than full cores was for safety, a secondary reason was
> that we don't have a way to treat them differently in the scheduler than
> 'full' cores and as a result were seeing that overall performance was
> often better when *not* using them.
I started to look at this last year, but didn't get very far because of
other projects.
Is anybody else actively working on the scheduler in this regard?
If not, I'll probably take another look at it. More and more CPUs that
would benefit from it are entering circulation now. My original motive
was due to the performance of the RK3399-based SBCs we've got here being
far below what it could be on OpenBSD, because processes tend to get
scheduled on one of the four slow cores, often leaving the two faster cores
completely idle.
But now I'm feeling the pain with our latest x86 machines too.
> And it seems that this second reason is unlikely to apply to Bulldozer.
Agreed.
> (A similar scheduler issue presents itself on systems with "big/little"
> cpu architectures on various ARM systems and P/E cores on newer Intel
> though of course in those cases they are separate cores).
Exactly.
No comments:
Post a Comment