On 26/10/2026 18:37, Tejun Heo wrote:
Hello, Tvrtko.

On Sat, Oct 24, 2026 at 09:33:42AM +0100, Tvrtko Ursulin wrote:
Can do. I'll get to it the week after next. (Tweaks to other patches in
the series as well.) Should I switch to a different base branch by then?

I just branched wq/for-7.5. That should work as the base.

But if you are curious how much better strict
affinity makes them I can measure, it is not a huge effort.

Yes, numbers with and without strict affinity would be great.

Re-measured submit latencies on three relevant setups:

    us     7.3.0   WQ_RT   +strict_affinity

   median    30      21          11
    min       2       3           2
    max     1038    941         108
   95pct    2384     49          28
   98pct    2922     63          51

I suppose we can interpret this in two ways. Plain WQ_RT is already a huge improvement, and although strict affinity is even better, the twice as good median probably isn't a huge selling point. So question is do tail latencies above the 98th percentile matter, or are they visible in the user experience. That I have no experience to answer.

FWIW I have locally split out the patch which adds strict affinity on top of WQ_RT so we can trivially go either way.

Hmm.. I could do that, but since the tool ships within the kernel does
it _need_ to have backward compatibility regarding kernel _internal_
implementation details?

It's just a nice-to-have, but other parts of the script already support
older kernels, so falling back would be good if it doesn't add too much
complexity.

Okay done.

Regards,

Tvrtko

Reply via email to