Re: [PATCH 2/6] mm: kswapd: Keep kswapd awake for high-orderallocations until a percentage of the node is balanced

[Mel Gorman]

On Fri, Dec 10, 2010 at 10:16:49AM +0900, KAMEZAWA Hiroyuki wrote:
> On Thu,  9 Dec 2010 11:18:16 +0000
> Mel Gorman <mel@xxxxxxxxx> wrote:
> 
> > When reclaiming for high-orders, kswapd is responsible for balancing a
> > node but it should not reclaim excessively. It avoids excessive reclaim by
> > considering if any zone in a node is balanced then the node is balanced. In
> > the cases where there are imbalanced zone sizes (e.g. ZONE_DMA with both
> > ZONE_DMA32 and ZONE_NORMAL), kswapd can go to sleep prematurely as just
> > one small zone was balanced.
> > 
> > This alters the sleep logic of kswapd slightly. It counts the number of pages
> > that make up the balanced zones. If the total number of balanced pages is
> > more than a quarter of the zone, kswapd will go back to sleep. This should
> > keep a node balanced without reclaiming an excessive number of pages.
> > 
> > Signed-off-by: Mel Gorman <mel@xxxxxxxxx>
> 
> Hmm, does this work well in
> 
> for example, x86-32,
> 	DMA: 16MB
> 	NORMAL: 700MB
> 	HIGHMEM: 11G
> ?
> 
> At 1st look, it's balanced when HIGHMEM has enough free pages...
> This is not good for NICs which requests high-order allocations.
> 

Good question.

In this case, the classzone_idx for the NICs high-order allocation will
be the Normal zone. In balance_pgdat(), this check is made

                                if (i <= classzone_idx)
                                        balanced += zone->present_pages;

Highmem will be too high and so the pages will not be counted and the node
will not be balanced.

> Can't we take claszone_idx into account at checking rather than
> node->present_pages ?
> 
> as
> 	balanced > present_pages_below_classzone_idx(node, classzone_idx)/4
> 
> ?

We can, but not for the reasons you list above. When a heavily
imbalanced highmem zone like this, the node might never be considered
balanced as the sum of DMA and Normal is less than 25% of the node.

-- 
Mel Gorman
Part-time Phd Student                          Linux Technology Center
University of Limerick                         IBM Dublin Software Lab
--
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/