ZVC/zone_reclaim: Leave 1% of unmapped pagecache pages for file I/O

ZVC/zone_reclaim: Leave 1% of unmapped pagecache pages for file I/O

[Christoph Lameter]

It turns out that it is advantageous to leave a small portion of
unmapped file backed pages if a zone is overallocated.

This allows recently used file I/O buffers to stay on the node and
reduces the times that zone reclaim is invoked if file I/O occurs
when we run out of memory in a zone.

Signed-off-by: Christoph Lameter <clameter@sgi.com>

Index: linux-2.6.17-mm4/mm/vmscan.c
===================================================================
--- linux-2.6.17-mm4.orig/mm/vmscan.c	2006-06-29 13:34:13.128150411 -0700
+++ linux-2.6.17-mm4/mm/vmscan.c	2006-06-29 19:44:54.717779791 -0700
@@ -1598,18 +1598,22 @@ int zone_reclaim(struct zone *zone, gfp_
 	int node_id;
 
 	/*
-	 * Do not reclaim if there are not enough reclaimable pages in this
-	 * zone that would satify this allocations.
+	 * Zone reclaim reclaims unmapped file backed pages.
 	 *
-	 * All unmapped pagecache pages are reclaimable.
+	 * A small portion of unmapped file backed pages is needed for
+	 * file I/O otherwise pages read by file I/O will be immediately
+	 * thrown out if the zone is overallocated. So we do not reclaim
+	 * if less than 1% of the zone is used by unmapped file backed pages.
 	 *
-	 * Both counters may be temporarily off a bit so we use
-	 * SWAP_CLUSTER_MAX as the boundary. It may also be good to
-	 * leave a few frequently used unmapped pagecache pages around.
+	 * The division by 128 approximates this and is here because a division
+	 * would be too expensive in this hot code path.
+	 *
+	 * Is it be useful to have a way to set the limit via /proc?
 	 */
 	if (zone_page_state(zone, NR_FILE_PAGES) -
-		zone_page_state(zone, NR_FILE_MAPPED) < SWAP_CLUSTER_MAX)
-			return 0;
+		zone_page_state(zone, NR_FILE_MAPPED) <
+			zone->present_pages / 128)
+				return 0;
 
 	/*
 	 * Avoid concurrent zone reclaims, do not reclaim in a zone that does

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: ZVC/zone_reclaim: Leave 1% of unmapped pagecache pages for file I/O

[Andrew Morton]

On Thu, 29 Jun 2006 19:51:38 -0700 (PDT)
Christoph Lameter <clameter@sgi.com> wrote:

> It turns out that it is advantageous to leave a small portion of
> unmapped file backed pages if a zone is overallocated.
>
> This allows recently used file I/O buffers to stay on the node and
> reduces the times that zone reclaim is invoked if file I/O occurs
> when we run out of memory in a zone.

I don't really understand this.  Can you expand? ie:

define "overallocated".

"turns out" how?  What problems were observed, and was was the behaviour
after the patch?

"reduces the times" from what down to what?

Thanks.

> Signed-off-by: Christoph Lameter <clameter@sgi.com>
> 
> Index: linux-2.6.17-mm4/mm/vmscan.c
> ===================================================================
> --- linux-2.6.17-mm4.orig/mm/vmscan.c	2006-06-29 13:34:13.128150411 -0700
> +++ linux-2.6.17-mm4/mm/vmscan.c	2006-06-29 19:44:54.717779791 -0700
> @@ -1598,18 +1598,22 @@ int zone_reclaim(struct zone *zone, gfp_
>  	int node_id;
>  
>  	/*
> -	 * Do not reclaim if there are not enough reclaimable pages in this
> -	 * zone that would satify this allocations.
> +	 * Zone reclaim reclaims unmapped file backed pages.
>  	 *
> -	 * All unmapped pagecache pages are reclaimable.
> +	 * A small portion of unmapped file backed pages is needed for
> +	 * file I/O otherwise pages read by file I/O will be immediately
> +	 * thrown out if the zone is overallocated. So we do not reclaim
> +	 * if less than 1% of the zone is used by unmapped file backed pages.
>  	 *
> -	 * Both counters may be temporarily off a bit so we use
> -	 * SWAP_CLUSTER_MAX as the boundary. It may also be good to
> -	 * leave a few frequently used unmapped pagecache pages around.
> +	 * The division by 128 approximates this and is here because a division
> +	 * would be too expensive in this hot code path.
> +	 *
> +	 * Is it be useful to have a way to set the limit via /proc?
>  	 */
>  	if (zone_page_state(zone, NR_FILE_PAGES) -
> -		zone_page_state(zone, NR_FILE_MAPPED) < SWAP_CLUSTER_MAX)
> -			return 0;
> +		zone_page_state(zone, NR_FILE_MAPPED) <
> +			zone->present_pages / 128)
> +				return 0;
>  
>  	/*
>  	 * Avoid concurrent zone reclaims, do not reclaim in a zone that does

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: ZVC/zone_reclaim: Leave 1% of unmapped pagecache pages for file I/O

[Christoph Lameter]

On Thu, 29 Jun 2006, Andrew Morton wrote:

> On Thu, 29 Jun 2006 19:51:38 -0700 (PDT)
> Christoph Lameter <clameter@sgi.com> wrote:
> 
> > It turns out that it is advantageous to leave a small portion of
> > unmapped file backed pages if a zone is overallocated.
> >
> > This allows recently used file I/O buffers to stay on the node and
> > reduces the times that zone reclaim is invoked if file I/O occurs
> > when we run out of memory in a zone.
> 
> I don't really understand this.  Can you expand? ie:
> 
> define "overallocated".

All pages (or almost all pages) in the zone are allocated and so the page 
allocator has to go off node.

> "turns out" how?  What problems were observed, and was was the behaviour
> after the patch?

The problem is that zone reclaim runs too frequently when the page cache 
is used for file I/O (read write and therefore unmapped pages!) alone and 
we have almost all pages of the zone allocated. Zone reclaim may remove 32 
unmapped pages. File I/O will use these pages for the next read/write 
requests and the unmapped pages increase. After the zone has filled up 
again zone reclaim will remove it again after only 32 pages. This cycle is 
too inefficient and there are potentially too many zone reclaim cycles.

With the 1% boundary we may still remove all unmapped pages for file 
I/O in zone reclaim pass. However. it will take a large number of read 
and writes to get back to 1% again where we trigger zone reclaim 
again.

The zone reclaim 2.6.16/17 does not show this behavior because
we have a 30 second timeout.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: ZVC/zone_reclaim: Leave 1% of unmapped pagecache pages for file I/O

[Andi Kleen]

On Friday 30 June 2006 04:51, Christoph Lameter wrote:
> It turns out that it is advantageous to leave a small portion of
> unmapped file backed pages if a zone is overallocated.
> 
> This allows recently used file I/O buffers to stay on the node and
> reduces the times that zone reclaim is invoked if file I/O occurs
> when we run out of memory in a zone.

Shouldn't that be some kind of tunable? Magic numbers are bad.

-Andi

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: ZVC/zone_reclaim: Leave 1% of unmapped pagecache pages for file I/O

[Christoph Lameter]

On Fri, 30 Jun 2006, Andi Kleen wrote:

> Shouldn't that be some kind of tunable? Magic numbers are bad.

Is this okay?

zone_reclaim: proc limit for the minimal amount of unmapped pagecache pages

Add /proc/sys/vm/min_unmapped to be able to control
the percentage of unmapped pages. Zone reclaim will only be triggered
if more than that number of unmapped pages exist in a zone.

And remove some outdated comments.

Signed-off-by: Christoph Lameter <clameter@sgi.com>

Index: linux-2.6.17-mm4/include/linux/swap.h
===================================================================
--- linux-2.6.17-mm4.orig/include/linux/swap.h	2006-06-29 13:34:12.572520730 -0700
+++ linux-2.6.17-mm4/include/linux/swap.h	2006-06-30 14:02:47.801113209 -0700
@@ -190,6 +190,7 @@ extern long vm_total_pages;
 
 #ifdef CONFIG_NUMA
 extern int zone_reclaim_mode;
+extern int sysctl_min_unmapped;
 extern int zone_reclaim(struct zone *, gfp_t, unsigned int);
 #else
 #define zone_reclaim_mode 0
Index: linux-2.6.17-mm4/mm/vmscan.c
===================================================================
--- linux-2.6.17-mm4.orig/mm/vmscan.c	2006-06-30 13:35:32.094681361 -0700
+++ linux-2.6.17-mm4/mm/vmscan.c	2006-06-30 14:04:48.663761285 -0700
@@ -1511,10 +1511,6 @@ module_init(kswapd_init)
  *
  * If non-zero call zone_reclaim when the number of free pages falls below
  * the watermarks.
- *
- * In the future we may add flags to the mode. However, the page allocator
- * should only have to check that zone_reclaim_mode != 0 before calling
- * zone_reclaim().
  */
 int zone_reclaim_mode __read_mostly;
 
@@ -1532,6 +1528,12 @@ int zone_reclaim_mode __read_mostly;
 #define ZONE_RECLAIM_PRIORITY 4
 
 /*
+ * Percentile of pages in a zone that must be unmapped
+ * for zone_reclaim to occur.
+ */
+int sysctl_min_unmapped = 1;
+
+/*
  * Try to free up some pages from this zone through reclaim.
  */
 static int __zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order)
@@ -1603,16 +1605,11 @@ int zone_reclaim(struct zone *zone, gfp_
 	 * A small portion of unmapped file backed pages is needed for
 	 * file I/O otherwise pages read by file I/O will be immediately
 	 * thrown out if the zone is overallocated. So we do not reclaim
-	 * if less than 1% of the zone is used by unmapped file backed pages.
-	 *
-	 * The division by 128 approximates this and is here because a division
-	 * would be too expensive in this hot code path.
-	 *
-	 * Is it be useful to have a way to set the limit via /proc?
+	 * if less than a specified percentage of the zone is used by
+	 * unmapped file backed pages.
 	 */
 	if (zone_page_state(zone, NR_FILE_PAGES) -
-		zone_page_state(zone, NR_FILE_MAPPED) <
-			zone->present_pages / 128)
+		zone_page_state(zone, NR_FILE_MAPPED) <= zone->min_unmapped)
 				return 0;
 
 	/*
Index: linux-2.6.17-mm4/mm/page_alloc.c
===================================================================
--- linux-2.6.17-mm4.orig/mm/page_alloc.c	2006-06-29 13:34:13.101784855 -0700
+++ linux-2.6.17-mm4/mm/page_alloc.c	2006-06-30 13:52:05.542017732 -0700
@@ -2098,6 +2098,10 @@ static void __meminit free_area_init_cor
 
 		zone->spanned_pages = size;
 		zone->present_pages = realsize;
+#ifdef CONFIG_NUMA
+		zone->min_unmapped = (realsize * sysctl_min_unmapped)
+						/ 100;
+#endif
 		zone->name = zone_names[j];
 		spin_lock_init(&zone->lock);
 		spin_lock_init(&zone->lru_lock);
@@ -2391,6 +2395,24 @@ int min_free_kbytes_sysctl_handler(ctl_t
 	return 0;
 }
 
+#ifdef CONFIG_NUMA
+int sysctl_min_unmapped_sysctl_handler(ctl_table *table, int write,
+	struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
+{
+	struct zone *zone;
+	int rc;
+
+	rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
+	if (rc)
+		return rc;
+
+	for_each_zone(zone)
+		zone->min_unmapped = (zone->present_pages *
+				sysctl_min_unmapped) / 100;
+	return 0;
+}
+#endif
+
 /*
  * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
  *	proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
Index: linux-2.6.17-mm4/kernel/sysctl.c
===================================================================
--- linux-2.6.17-mm4.orig/kernel/sysctl.c	2006-06-29 13:34:12.980698597 -0700
+++ linux-2.6.17-mm4/kernel/sysctl.c	2006-06-30 13:50:32.609293703 -0700
@@ -998,6 +998,17 @@ static ctl_table vm_table[] = {
 		.strategy	= &sysctl_intvec,
 		.extra1		= &zero,
 	},
+	{
+		.ctl_name	= VM_MIN_UNMAPPED,
+		.procname	= "min_unmapped",
+		.data		= &sysctl_min_unmapped,
+		.maxlen		= sizeof(sysctl_min_unmapped),
+		.mode		= 0644,
+		.proc_handler	= &sysctl_min_unmapped_sysctl_handler,
+		.strategy	= &sysctl_intvec,
+		.extra1		= &zero,
+		.extra2		= &one_hundred,
+	},
 #endif
 #ifdef CONFIG_X86_32
 	{
Index: linux-2.6.17-mm4/include/linux/sysctl.h
===================================================================
--- linux-2.6.17-mm4.orig/include/linux/sysctl.h	2006-06-29 13:34:12.603768796 -0700
+++ linux-2.6.17-mm4/include/linux/sysctl.h	2006-06-30 13:50:32.610270205 -0700
@@ -189,7 +189,7 @@ enum
 	VM_DROP_PAGECACHE=29,	/* int: nuke lots of pagecache */
 	VM_PERCPU_PAGELIST_FRACTION=30,/* int: fraction of pages in each percpu_pagelist */
 	VM_ZONE_RECLAIM_MODE=31, /* reclaim local zone memory before going off node */
-	VM_ZONE_RECLAIM_INTERVAL=32, /* time period to wait after reclaim failure */
+	VM_MIN_UNMAPPED=32,	/* Set min percent of unmapped pages */
 	VM_PANIC_ON_OOM=33,	/* panic at out-of-memory */
 	VM_VDSO_ENABLED=34,	/* map VDSO into new processes? */
 	VM_SWAP_PREFETCH=35,	/* swap prefetch */
Index: linux-2.6.17-mm4/include/linux/mmzone.h
===================================================================
--- linux-2.6.17-mm4.orig/include/linux/mmzone.h	2006-06-29 13:34:12.302029655 -0700
+++ linux-2.6.17-mm4/include/linux/mmzone.h	2006-06-30 13:50:32.611246707 -0700
@@ -150,6 +150,10 @@ struct zone {
 	unsigned long		lowmem_reserve[MAX_NR_ZONES];
 
 #ifdef CONFIG_NUMA
+	/*
+	 * zone reclaim becomes active if more unmapped pages exist.
+	 */
+	unsigned long		min_unmapped;
 	struct per_cpu_pageset	*pageset[NR_CPUS];
 #else
 	struct per_cpu_pageset	pageset[NR_CPUS];
@@ -420,6 +424,8 @@ int lowmem_reserve_ratio_sysctl_handler(
 					void __user *, size_t *, loff_t *);
 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int, struct file *,
 					void __user *, size_t *, loff_t *);
+int sysctl_min_unmapped_sysctl_handler(struct ctl_table *, int,
+			struct file *, void __user *, size_t *, loff_t *);
 
 #include <linux/topology.h>
 /* Returns the number of the current Node. */
Index: linux-2.6.17-mm4/Documentation/sysctl/vm.txt
===================================================================
--- linux-2.6.17-mm4.orig/Documentation/sysctl/vm.txt	2006-06-29 13:34:05.678416074 -0700
+++ linux-2.6.17-mm4/Documentation/sysctl/vm.txt	2006-06-30 13:50:32.611246707 -0700
@@ -28,6 +28,7 @@ Currently, these files are in /proc/sys/
 - block_dump
 - drop-caches
 - zone_reclaim_mode
+- min_unmapped
 - panic_on_oom
 - swap_prefetch
 - readahead_ratio
@@ -171,6 +172,17 @@ in all nodes of the system.
 
 =============================================================
 
+min_unmapped:
+
+A percentage of the file backed pages in each zone. Zone reclaim will only
+occur if more than this percentage of pages are file backed and unmapped.
+This is to insure that a minimal amount of local pages is still available
+for file I/O even if the node is overallocated.
+
+The default is 1 percent.
+
+=============================================================
+
 panic_on_oom
 
 This enables or disables panic on out-of-memory feature.  If this is set to 1,

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: ZVC/zone_reclaim: Leave 1% of unmapped pagecache pages for file I/O

[Andrew Morton]

Christoph Lameter <clameter@sgi.com> wrote:
>
> +min_unmapped:
> +
> +A percentage of the file backed pages in each zone. Zone reclaim will only
> +occur if more than this percentage of pages are file backed and unmapped.
> +This is to insure that a minimal amount of local pages is still available
> +for file I/O even if the node is overallocated.
> +
> +The default is 1 percent.
> +

Probably this should have a big "NUMA only" slapped on it so people don't
wonder why they don't have such a file.

Also, it's nice if the naming of these control files can communicate the
units to the operator.  In the case of /proc/sys/vm we use _ratio to
indicate that it's a proportional control in the 0% - 100% range.

So /proc/sys/vm/min_unmapped_ratio would be clearer, and more consistent.

<looks forward to the day when this becomes per-node :(>

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>