Re: [PATCH 0/2] convert mmap_sem to a scalable rw_mutex

[Peter Zijlstra]

On Fri, 2007-05-11 at 18:52 +0200, Eric Dumazet wrote:
> Ingo Molnar a Ãcrit :
> > * Peter Zijlstra <a.p.zijlstra@xxxxxxxxx> wrote:
> > 
> >> I was toying with a scalable rw_mutex and found that it gives ~10% 
> >> reduction in system time on ebizzy runs (without the MADV_FREE patch).
> >>
> >> 2-way x86_64 pentium D box:
> >>
> >> 2.6.21
> >>
> >> /usr/bin/time ./ebizzy -m -P
> >> 59.49user 137.74system 1:49.22elapsed 180%CPU (0avgtext+0avgdata 0maxresident)k
> >> 0inputs+0outputs (0major+33555877minor)pagefaults 0swaps
> >>
> >> 2.6.21-rw_mutex
> >>
> >> /usr/bin/time ./ebizzy -m -P
> >> 57.85user 124.30system 1:42.99elapsed 176%CPU (0avgtext+0avgdata 0maxresident)k
> >> 0inputs+0outputs (0major+33555877minor)pagefaults 0swaps
> > 
> > nice! This 6% runtime reduction on a 2-way box will i suspect get 
> > exponentially better on systems with more CPUs/cores.
> 
> As long you only have readers, yes.
> 
> But I personally find this new rw_mutex not scalable at all if you have some 
> writers around.
> 
> percpu_counter_sum is just a L1 cache eater, and O(NR_CPUS)

Yeah, that is true; there are two occurences, the one in
rw_mutex_read_unlock() is not strictly needed for correctness.

Write locks are indeed quite expensive. But given the ratio of
reader:writer locks on mmap_sem (I'm not all that familiar with other
rwsem users) this trade-off seems workable.

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/