Re: [PATCH v7 00/13] fold per-CPU vmstats remotely

[Marcelo Tosatti <mtosatti@redhat.com>]

On Thu, Apr 27, 2023 at 10:39:29AM +0200, Michal Hocko wrote:
> On Wed 26-04-23 13:10:54, Marcelo Tosatti wrote:
> [...]
> > "To test the performance difference, a page allocator microbenchmark:
> > https://github.com/netoptimizer/prototype-kernel/blob/master/kernel/mm/bench/page_bench01.c
> > with loops=1000000 was used, on Intel Core i7-11850H @ 2.50GHz.
> > 
> > For the single_page_alloc_free test, which does
> > 
> >        	/** Loop to measure **/
> >        	for (i = 0; i < rec->loops; i++) {
> >                	my_page = alloc_page(gfp_mask);
> >                 if (unlikely(my_page == NULL))
> >                        	return 0;
> >                 __free_page(my_page);
> >         }                                                                                                           
> > 
> > Unit is cycles.
> > 
> > Vanilla                 Patched         Diff
> > 115.25                  117             1.4%"
> > 
> > To be honest, that 1.4% difference was not stable but fluctuated between
> > positive and negative percentages (so the performance difference was in
> > the noise).
> > 
> > So performance is not a decisive factor in this case.
> 
> It is not neglible considering that majority worklods will not benefit
> from this change. You are clearly ignoring that vmstat code has been
> highly optimized for local per-cpu access exactly to avoid locked
> operations and cache line bouncing.
> -- 
> Michal Hocko
> SUSE Labs

Again, the values fluctuate between positive and negative
performance difference (i happen to have copied a positive value).

So the performance difference is in the noise (its not stable at 1.4%),
but rather close to 0%.

So the data is showing that there is no negative performance impact.