[I do not have time to get over all points here and will be offline
until Monday - will get back to the rest then]
On Tue 07-10-14 21:11:06, Johannes Weiner wrote:
> On Tue, Oct 07, 2014 at 03:59:50PM +0200, Michal Hocko wrote:
[...]
> > I am completely missing any notes about potential excessive
> > swapouts or longer reclaim stalls which are a natural side effect of direct
> > reclaim with a larger target (or is this something we do not agree on?).
>
> Yes, we disagree here. Why is reclaiming 2MB once worse than entering
> reclaim 16 times to reclaim SWAP_CLUSTER_MAX?
You can enter DEF_PRIORITY reclaim 16 times and reclaim your target but
you need at least 512<<DEF_PRIORITY pages on your LRUs to do it in a
single run on that priority. So especially small groups will pay more
and would be subject to mentioned problems (e.g. over-reclaim).
> There is no inherent difference in reclaiming a big chunk and
> reclaiming many small chunks that add up to the same size.
[...]
> > Another part that matters is the size. Memcgs might be really small and
> > that changes the math. Large reclaim target will get to low prio reclaim
> > and thus the excessive reclaim.
>
> I already addressed page size vs. memcg size before.
>
> However, low priority reclaim does not result in excessive reclaim.
> The reclaim goal is checked every time it scanned SWAP_CLUSTER_MAX
> pages, and it exits if the goal has been met. See shrink_lruvec(),
> shrink_zone() etc.
Now I am confused. shrink_zone will bail out but shrink_lruvec will loop
over nr[...] until they are empty and only updates the numbers to be
roughly proportional once:
if (nr_reclaimed < nr_to_reclaim || scan_adjusted)
continue;
/*
* For kswapd and memcg, reclaim at least the number of pages
* requested. Ensure that the anon and file LRUs are scanned
* proportionally what was requested by get_scan_count(). We
* stop reclaiming one LRU and reduce the amount scanning
* proportional to the original scan target.
*/
[...]
scan_adjusted = true;
Or do you rely on
/*
* It's just vindictive to attack the larger once the smaller
* has gone to zero. And given the way we stop scanning the
* smaller below, this makes sure that we only make one nudge
* towards proportionality once we've got nr_to_reclaim.
*/
if (!nr_file || !nr_anon)
break;
and SCAN_FILE because !inactive_file_is_low?
[...]
--
Michal Hocko
SUSE Labs
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>