On Wed, 2008-08-13 at 23:14 +0900, KOSAKI Motohiro wrote:
> >> :t-0000128 28739 128 1.3G 20984/20984/8 512 0 99 0 *
> >
> > Argh. Most slabs contain a single object. Probably due to the conflict resolution.
>
> agreed with the issue exist in lock contention code.
>
>
> > The obvious fix is to avoid allocating another slab on conflict but how will
> > this impact performance?
> >
> >
> > Index: linux-2.6/mm/slub.c
> > ===================================================================
> > --- linux-2.6.orig/mm/slub.c 2008-08-13 08:06:00.000000000 -0500
> > +++ linux-2.6/mm/slub.c 2008-08-13 08:07:59.000000000 -0500
> > @@ -1253,13 +1253,11 @@
> > static inline int lock_and_freeze_slab(struct kmem_cache_node *n,
> > struct page *page)
> > {
> > - if (slab_trylock(page)) {
> > - list_del(&page->lru);
> > - n->nr_partial--;
> > - __SetPageSlubFrozen(page);
> > - return 1;
> > - }
> > - return 0;
> > + slab_lock(page);
> > + list_del(&page->lru);
> > + n->nr_partial--;
> > + __SetPageSlubFrozen(page);
> > + return 1;
> > }
>
> I don't mesure it yet. I don't like this patch.
> maybe, it decrease other typical benchmark.
>
> So, I think better way is
>
> 1. slab_trylock(), if success goto 10.
> 2. check fragmentation ratio, if low goto 10
> 3. slab_lock()
> 10. return func
>
> I think this way doesn't cause performance regression.
> because high fragmentation cause defrag and compaction lately.
> So, prevent fragmentation often increase performance.
>
> Thought?
I guess that would work. But how exactly would you quantify
"fragmentation ratio?"
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
