Re: [PATCH] mm/compaction: restore COMPACT_CLUSTER_MAX to 32

[Mel Gorman]

On Mon, Dec 07, 2015 at 05:00:41PM -0800, Andrew Morton wrote:
> On Thu,  3 Dec 2015 13:11:40 +0900 Joonsoo Kim  wrote:
> 
> > Until now, COMPACT_CLUSTER_MAX is defined as SWAP_CLUSTER_MAX.
> > Commit ("mm: increase SWAP_CLUSTER_MAX to batch TLB flushes")
> > changes SWAP_CLUSTER_MAX from 32 to 256 to improve tlb flush performance
> > so COMPACT_CLUSTER_MAX is also changed to 256.
> 
> "mm: increase SWAP_CLUSTER_MAX to batch TLB flushes" has been in limbo
> for quite a while.  Because it has been unclear whether the patch's
> benefits exceed its costs+risks.
> 
> We should make a decision here - either do the appropriate testing or
> drop the patch.
> 

At this point, drop it. The benefits that apply to some corner cases are
marginal but the concerns about potentially isolating and reclaiming too
much persist.

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

Re: [PATCH] mm/compaction: restore COMPACT_CLUSTER_MAX to 32

[Mel Gorman]

On Mon, Dec 07, 2015 at 05:00:41PM -0800, Andrew Morton wrote:
> On Thu,  3 Dec 2015 13:11:40 +0900 Joonsoo Kim  wrote:
> 
> > Until now, COMPACT_CLUSTER_MAX is defined as SWAP_CLUSTER_MAX.
> > Commit ("mm: increase SWAP_CLUSTER_MAX to batch TLB flushes")
> > changes SWAP_CLUSTER_MAX from 32 to 256 to improve tlb flush performance
> > so COMPACT_CLUSTER_MAX is also changed to 256.
> 
> "mm: increase SWAP_CLUSTER_MAX to batch TLB flushes" has been in limbo
> for quite a while.  Because it has been unclear whether the patch's
> benefits exceed its costs+risks.
> 
> We should make a decision here - either do the appropriate testing or
> drop the patch.
> 

At this point, drop it. The benefits that apply to some corner cases are
marginal but the concerns about potentially isolating and reclaiming too
much persist.

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

Re: [PATCH] mm/compaction: restore COMPACT_CLUSTER_MAX to 32

[Andrew Morton]

On Thu,  3 Dec 2015 13:11:40 +0900 Joonsoo Kim  wrote:

> Until now, COMPACT_CLUSTER_MAX is defined as SWAP_CLUSTER_MAX.
> Commit ("mm: increase SWAP_CLUSTER_MAX to batch TLB flushes")
> changes SWAP_CLUSTER_MAX from 32 to 256 to improve tlb flush performance
> so COMPACT_CLUSTER_MAX is also changed to 256.

"mm: increase SWAP_CLUSTER_MAX to batch TLB flushes" has been in limbo
for quite a while.  Because it has been unclear whether the patch's
benefits exceed its costs+risks.

We should make a decision here - either do the appropriate testing or
drop the patch.


> But, it has
> no justification on compaction-side and I think that loss is more than
> benefit.
> 
> One example is that migration scanner would isolates and migrates
> too many pages unnecessarily with 256 COMPACT_CLUSTER_MAX. It may be
> enough to migrate 4 pages in order to make order-2 page, but, now,
> compaction will migrate 256 pages.
> 
> To reduce this unneeded overhead, this patch restores
> COMPACT_CLUSTER_MAX to 32.
> 
> ...
>
> --- a/include/linux/swap.h
> +++ b/include/linux/swap.h
> @@ -155,7 +155,7 @@ enum {
>  };
>  
>  #define SWAP_CLUSTER_MAX 256UL
> -#define COMPACT_CLUSTER_MAX SWAP_CLUSTER_MAX
> +#define COMPACT_CLUSTER_MAX 32UL
>  
>  /*
>   * Ratio between zone->managed_pages and the "gap" that above the per-zone

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

Re: [PATCH] mm/compaction: restore COMPACT_CLUSTER_MAX to 32

[Andrew Morton]

On Thu,  3 Dec 2015 13:11:40 +0900 Joonsoo Kim  wrote:

> Until now, COMPACT_CLUSTER_MAX is defined as SWAP_CLUSTER_MAX.
> Commit ("mm: increase SWAP_CLUSTER_MAX to batch TLB flushes")
> changes SWAP_CLUSTER_MAX from 32 to 256 to improve tlb flush performance
> so COMPACT_CLUSTER_MAX is also changed to 256.

"mm: increase SWAP_CLUSTER_MAX to batch TLB flushes" has been in limbo
for quite a while.  Because it has been unclear whether the patch's
benefits exceed its costs+risks.

We should make a decision here - either do the appropriate testing or
drop the patch.


> But, it has
> no justification on compaction-side and I think that loss is more than
> benefit.
> 
> One example is that migration scanner would isolates and migrates
> too many pages unnecessarily with 256 COMPACT_CLUSTER_MAX. It may be
> enough to migrate 4 pages in order to make order-2 page, but, now,
> compaction will migrate 256 pages.
> 
> To reduce this unneeded overhead, this patch restores
> COMPACT_CLUSTER_MAX to 32.
> 
> ...
>
> --- a/include/linux/swap.h
> +++ b/include/linux/swap.h
> @@ -155,7 +155,7 @@ enum {
>  };
>  
>  #define SWAP_CLUSTER_MAX 256UL
> -#define COMPACT_CLUSTER_MAX SWAP_CLUSTER_MAX
> +#define COMPACT_CLUSTER_MAX 32UL
>  
>  /*
>   * Ratio between zone->managed_pages and the "gap" that above the per-zone

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

[PATCH] mm/compaction: restore COMPACT_CLUSTER_MAX to 32

[Joonsoo Kim]

Until now, COMPACT_CLUSTER_MAX is defined as SWAP_CLUSTER_MAX.
Commit ("mm: increase SWAP_CLUSTER_MAX to batch TLB flushes")
changes SWAP_CLUSTER_MAX from 32 to 256 to improve tlb flush performance
so COMPACT_CLUSTER_MAX is also changed to 256. But, it has
no justification on compaction-side and I think that loss is more than
benefit.

One example is that migration scanner would isolates and migrates
too many pages unnecessarily with 256 COMPACT_CLUSTER_MAX. It may be
enough to migrate 4 pages in order to make order-2 page, but, now,
compaction will migrate 256 pages.

To reduce this unneeded overhead, this patch restores
COMPACT_CLUSTER_MAX to 32.

Signed-off-by: Joonsoo Kim 
---
 include/linux/swap.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/include/linux/swap.h b/include/linux/swap.h
index d08feef..31eb343 100644
--- a/include/linux/swap.h
+++ b/include/linux/swap.h
@@ -155,7 +155,7 @@ enum {
 };
 
 #define SWAP_CLUSTER_MAX 256UL
-#define COMPACT_CLUSTER_MAX SWAP_CLUSTER_MAX
+#define COMPACT_CLUSTER_MAX 32UL
 
 /*
  * Ratio between zone->managed_pages and the "gap" that above the per-zone
-- 
1.9.1

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

[PATCH] mm/compaction: restore COMPACT_CLUSTER_MAX to 32

[Joonsoo Kim]

Until now, COMPACT_CLUSTER_MAX is defined as SWAP_CLUSTER_MAX.
Commit ("mm: increase SWAP_CLUSTER_MAX to batch TLB flushes")
changes SWAP_CLUSTER_MAX from 32 to 256 to improve tlb flush performance
so COMPACT_CLUSTER_MAX is also changed to 256. But, it has
no justification on compaction-side and I think that loss is more than
benefit.

One example is that migration scanner would isolates and migrates
too many pages unnecessarily with 256 COMPACT_CLUSTER_MAX. It may be
enough to migrate 4 pages in order to make order-2 page, but, now,
compaction will migrate 256 pages.

To reduce this unneeded overhead, this patch restores
COMPACT_CLUSTER_MAX to 32.

Signed-off-by: Joonsoo Kim 
---
 include/linux/swap.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/include/linux/swap.h b/include/linux/swap.h
index d08feef..31eb343 100644
--- a/include/linux/swap.h
+++ b/include/linux/swap.h
@@ -155,7 +155,7 @@ enum {
 };
 
 #define SWAP_CLUSTER_MAX 256UL
-#define COMPACT_CLUSTER_MAX SWAP_CLUSTER_MAX
+#define COMPACT_CLUSTER_MAX 32UL
 
 /*
  * Ratio between zone->managed_pages and the "gap" that above the per-zone
-- 
1.9.1

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