lkml.org 
[lkml]   [2012]   [Sep]   [24]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH v3 06/16] memcg: infrastructure to match an allocation to the right cache
Hello, Glauber.

On Mon, Sep 24, 2012 at 12:46:35PM +0400, Glauber Costa wrote:
> >> +#ifdef CONFIG_MEMCG_KMEM
> >> + /* Slab accounting */
> >> + struct kmem_cache *slabs[MAX_KMEM_CACHE_TYPES];
> >> +#endif
> >
> > Bah, 400 entry array in struct mem_cgroup. Can't we do something a
> > bit more flexible?
> >
>
> I guess. I still would like it to be an array, so we can easily access
> its fields. There are two ways around this:
>
> 1) Do like the events mechanism and allocate this in a separate
> structure. Add a pointer chase in the access, and I don't think it helps
> much because it gets allocated anyway. But we could at least
> defer it to the time when we limit the cache.

Start at some reasonable size and then double it as usage grows? How
many kmem_caches do we typically end up using?

> >> + if (memcg->slabs[idx] == NULL) {
> >> + memcg_create_cache_enqueue(memcg, cachep);
> >
> > Do we want to wait for the work item if @gfp allows?
> >
>
> I tried this once, and it got complicated enough that I deemed as "not
> worth it". I honestly don't remember much of the details now, it was one
> of the first things I tried, and a bunch of time has passed. If you
> think it is absolutely worth it, I can try it again. But at the very
> best, I view this as an optimization.

I don't know. It seems like a logical thing to try and depends on how
complex it gets. I don't think it's a must. The whole thing is
somewhat opportunistic after all.

Thanks.

--
tejun


\
 
 \ /
  Last update: 2012-09-24 20:21    [W:0.079 / U:0.088 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site