Messages in this thread | | | Subject | Re: [x86/signal] 3aac3ebea0: will-it-scale.per_thread_ops -11.9% regression | From | Dave Hansen <> | Date | Tue, 7 Dec 2021 15:14:38 -0800 |
| |
On 12/6/21 5:21 PM, kernel test robot wrote: > > 1bdda24c4af64cd2 3aac3ebea08f2d342364f827c89 > ---------------- --------------------------- > %stddev %change %stddev > \ | \ > 980404 ± 3% -10.2% 880436 ± 2% will-it-scale.16.threads > 61274 ± 3% -10.2% 55027 ± 2% will-it-scale.per_thread_ops > 980404 ± 3% -10.2% 880436 ± 2% will-it-scale.workload > 9745749 ± 18% +26.8% 12356608 ± 4% meminfo.DirectMap2M
Something else funky is going on here. Why would there all of a sudden be so many more 2M pages in the direct map? I also see gunk like interrupts on the network card going up. I can certainly see that happening if something else on the network was messing around.
Granted, this was seen across several systems, but it's really odd. I guess I'll go try to dig up one of the actual ones where this was seen.
I tried on a smaller Skylake system and I don't see any regression at all or any interesting delta in a perf profile.
Oliver or Chang, could you try to reproduce this by hand on one of the suspect systems? Build:
1bdda24c4a ("signal: Add an optional check for altstack size")
then run will-it-scale by hand. Then build:
3aac3ebea0 ("x86/signal: Implement sigaltstack size validation")
and run it again. Also, do we see any higher core-count regressions? These all seem to happen with:
mode=thread nr_task=16
That's really odd to see that for these systems with probably ~50 cores each. I'd expect to see it get worse at higher core counts.
| |