Re: [x86/mce] 7bb39313cd: netperf.Throughput_tps -4.5% regression

[Feng Tang]

On Tue, Jan 19, 2021 at 04:33:50PM +0100, Borislav Petkov wrote:
> On Tue, Jan 19, 2021 at 11:09:03PM +0800, Feng Tang wrote:
> > Yes, that can happen. I started a 4 tasks netperf on a 4C/8T KBL desktop,
> > and also saw around 2% improvement. Both the kernel config and the
> > platform matters.
> 
> Oh great. ;-\
> 
> > For the performance changes I have checked, sometimes the change can be
> > reproduced on platforms of different generations (the exact delta number
> > may differs), sometimes it can only be reproduced on one specific platform,
> > like some old generation, or special one like Xeon Phi.
> 
> Probably because that Xeon Phi thing is not as powerful cache-wise as
> some newer ones which have bigger caches and smarter hw prefetchers,
> etc.

Yes, cache size/architecture/policy plays a critical role in these
benchmarking, which is our first thing to check for these strange
kernel performance changes.

Thanks,
Feng