Re: [PATCH mmotm 2.5/4] memcg: disable irq at page cgroup lock (Re: [PATCH -mmotm 3/4] memcg: dirty pages accounting and limiting infrastructure)

[Balbir Singh <balbir-23VcF4HTsmIX0ybBhKVfKdBPR1lH4CV8@public.gmane.org>]

* nishimura-YQH0OdQVrdy45+QrQBaojngSJqDPrsil@public.gmane.org <nishimura-YQH0OdQVrdy45+QrQBaojngSJqDPrsil@public.gmane.org> [2010-03-10 10:43:09]:

> > Please please measure the performance overhead of this change.
> > 
> 
> here.
> 
> > > > > > > I made a patch below and measured the time(average of 10 times) of kernel build
> > > > > > > on tmpfs(make -j8 on 8 CPU machine with 2.6.33 defconfig).
> > > > > > > 
> > > > > > > <before>
> > > > > > > - root cgroup: 190.47 sec
> > > > > > > - child cgroup: 192.81 sec
> > > > > > > 
> > > > > > > <after>
> > > > > > > - root cgroup: 191.06 sec
> > > > > > > - child cgroup: 193.06 sec
> > > > > > > 
> 
> <after2(local_irq_save/restore)>
> - root cgroup: 191.42 sec
> - child cgroup: 193.55 sec
> 
> hmm, I think it's in error range, but I can see a tendency by testing several times
> that it's getting slower as I add additional codes. Using local_irq_disable()/enable()
> except in mem_cgroup_update_file_mapped(it can be the only candidate to be called
> with irq disabled in future) might be the choice.
>

Error range would depend on things like standard deviation and
repetition. It might be good to keep update_file_mapped and see the
impact. My concern is with large systems, the difference might be
larger.
 
-- 
	Three Cheers,
	Balbir