lkml.org 
[lkml]   [2010]   [Aug]   [30]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
SubjectRe: [PATCH] vmscan: prevent background aging of anon page in no swap system
Date
> On Sun, Aug 29, 2010 at 5:18 PM, Minchan Kim <minchan.kim@gmail.com> wrote:
> > Hi Ying,
> >
> > On Mon, Aug 30, 2010 at 6:23 AM, Ying Han <yinghan@google.com> wrote:
> >> On Sun, Aug 29, 2010 at 1:03 PM, Rik van Riel <riel@redhat.com> wrote:
> >>> On 08/29/2010 01:45 PM, Ying Han wrote:
> >>>
> >>>> There are few other places in vmscan where we check nr_swap_pages and
> >>>> inactive_anon_is_low. Are we planning to change them to use
> >>>> total_swap_pages
> >>>> to be consistent ?
> >>>
> >>> If that makes sense, maybe the check can just be moved into
> >>> inactive_anon_is_low itself?
> >>
> >> That was the initial patch posted, instead we changed to use
> >> total_swap_pages instead. How this patch looks:
> >>
> >> @@ -1605,6 +1605,9 @@ static int inactive_anon_is_low(struct zone
> >> *zone, struct scan_control *sc)
> >>  {
> >>        int low;
> >>
> >> +       if (total_swap_pages <= 0)
> >> +               return 0;
> >> +
> >>        if (scanning_global_lru(sc))
> >>                low = inactive_anon_is_low_global(zone);
> >>        else
> >> @@ -1856,7 +1859,7 @@ static void shrink_zone(int priority, struct zone *zone,
> >>         * Even if we did not try to evict anon pages at all, we want to
> >>         * rebalance the anon lru active/inactive ratio.
> >>         */
> >> -       if (inactive_anon_is_low(zone, sc) && nr_swap_pages > 0)
> >> +       if (inactive_anon_is_low(zone, sc))
> >>                shrink_active_list(SWAP_CLUSTER_MAX, zone, sc, priority, 0);
> >>
> >>        throttle_vm_writeout(sc->gfp_mask);
> >>
> >> --Ying
> >>
> >>>
> >
> > I did it intentionally since inactive_anon_is_low have been used both
> > direct reclaim and background path. In this point, your patch could
> > make side effect in swap enabled system when swap is full.
> >
> > I think we need aging in only background if system is swap full.
> > That's because if the swap space is full, we don't reclaim anon pages
> > in direct reclaim path with (nr_swap_pages < 0)  and even have been
> > not rebalance it until now.
> > I think direct reclaim path is important about latency as well as
> > reclaim's effectiveness.
> > So if you don't mind, I hope direct reclaim patch would be left just as it is.
>
> Minchan, I would prefer to make kswapd as well as direct reclaim to be
> consistent if possible.
> They both try to reclaim pages when system is under memory pressure,
> and also do not make
> much sense to look at anon lru if no swap space available. Either
> because of no swapon or run
> out of swap space.
>
> I think letting kswapd to age anon lru without free swap space is not
> necessary neither. That leads
> to my initial patch:
>
> @@ -1605,6 +1605,9 @@ static int inactive_anon_is_low(struct zone
> *zone, struct scan_control *sc)
> {
> int low;
>
> + if (nr_swap_pages <= 0)
> + return 0;
> +
> if (scanning_global_lru(sc))
> low = inactive_anon_is_low_global(zone);
> else
> @@ -1856,7 +1859,7 @@ static void shrink_zone(int priority, struct zone *zone,
> * Even if we did not try to evict anon pages at all, we want to
> * rebalance the anon lru active/inactive ratio.
> */
> - if (inactive_anon_is_low(zone, sc) && nr_swap_pages > 0)
> + if (inactive_anon_is_low(zone, sc))
> shrink_active_list(SWAP_CLUSTER_MAX, zone, sc, priority, 0);
>
> What do you think ?

Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>


I think both Ying's and Minchan's opnion are right and makes sense. however I _personally_
like Ying version because 1) this version is simpler 2) swap full is very rarely event 3)
no swap mounting is very common on HPC. so this version could have a chance to
improvement hpc workload too.

In the other word, both avoiding unnecessary TLB flush and keeping proper page aging are
performance matter. so when we are talking performance, we always need to think frequency
of the event.

Anyway I'm very glad minchan who embedded developer pay attention server workload
carefully. Very thanks.




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

\
 
 \ /
  Last update: 2010-08-31 02:59    [W:0.083 / U:0.272 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site