Optimize 

Index: linux-2.6.18-rc4/mm/vmstat.c
===================================================================
--- linux-2.6.18-rc4.orig/mm/vmstat.c	2006-08-18 14:52:03.478999664 -0700
+++ linux-2.6.18-rc4/mm/vmstat.c	2006-08-18 15:14:06.191296910 -0700
@@ -168,17 +168,13 @@ static s8 calculate_threshold(struct zon
 static void refresh_zone_stat_thresholds(void)
 {
 	struct zone *zone;
-	int cpu;
-	int threshold;
 
 	for_each_zone(zone) {
 		if (!zone->present_pages)
 			continue;
 
-		threshold = calculate_threshold(zone);
-
-		for_each_online_cpu(cpu)
-			zone_pcp(zone, cpu)->stat_threshold = threshold;
+		zone->stat_threshold =
+			calculate_threshold(zone);
 	}
 }
 
@@ -194,7 +190,7 @@ void __mod_zone_page_state(struct zone *
 
 	x = delta + *p;
 
-	if (unlikely(x > pcp->stat_threshold || x < -pcp->stat_threshold)) {
+	if (unlikely(x > zone->stat_threshold || x < -zone->stat_threshold)) {
 		zone_page_state_add(x, zone, item);
 		x = 0;
 	}
@@ -244,8 +240,8 @@ static void __inc_zone_state(struct zone
 
 	(*p)++;
 
-	if (unlikely(*p > pcp->stat_threshold)) {
-		int overstep = pcp->stat_threshold / 2;
+	if (unlikely(*p > zone->stat_threshold)) {
+		int overstep = zone->stat_threshold / 2;
 
 		zone_page_state_add(*p + overstep, zone, item);
 		*p = - overstep;
@@ -266,8 +262,8 @@ void __dec_zone_page_state(struct page *
 
 	(*p)--;
 
-	if (unlikely(*p < - pcp->stat_threshold)) {
-		int overstep = pcp->stat_threshold / 2;
+	if (unlikely(*p < - zone->stat_threshold)) {
+		int overstep = zone->stat_threshold / 2;
 
 		zone_page_state_add(*p - overstep, zone, item);
 		*p = overstep;
@@ -312,7 +308,7 @@ EXPORT_SYMBOL(dec_zone_page_state);
 void refresh_cpu_vm_stats(int cpu)
 {
 	struct zone *zone;
-	int i;
+	int i,j;
 	unsigned long flags;
 
 	for_each_zone(zone) {
@@ -320,14 +316,25 @@ void refresh_cpu_vm_stats(int cpu)
 
 		pcp = zone_pcp(zone, cpu);
 
-		for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
-			if (pcp->vm_stat_diff[i]) {
+		for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i += sizeof(long)) {
+			/*
+			 * Use a long fetch to check a group of counters
+			 * at a time. This works because the pcp structure
+			 * is cacheline aligned. Extraneous counters at the
+			 * end are always zero.
+			 */
+			if (! *(long *)(pcp->vm_stat_diff + i))
+				continue;
+
+			for(j = i; j < i + sizeof(long); j++)
+			if (pcp->vm_stat_diff[j]) {
 				local_irq_save(flags);
-				zone_page_state_add(pcp->vm_stat_diff[i],
-					zone, i);
-				pcp->vm_stat_diff[i] = 0;
+				zone_page_state_add(pcp->vm_stat_diff[j],
+					zone, j);
+				pcp->vm_stat_diff[j] = 0;
 				local_irq_restore(flags);
 			}
+		}
 	}
 }
 
@@ -575,8 +582,6 @@ static int zoneinfo_show(struct seq_file
 					   pageset->pcp[j].high,
 					   pageset->pcp[j].batch);
 			}
-			seq_printf(m, "\n    vm stats threshold: %d",
-						pageset->stat_threshold);
 		}
 		seq_printf(m,
 			   "\n  all_unreclaimable: %u"
@@ -588,7 +593,8 @@ static int zoneinfo_show(struct seq_file
 			   zone->temp_priority,
 			   zone->zone_start_pfn);
 		spin_unlock_irqrestore(&zone->lock, flags);
-		seq_putc(m, '\n');
+		seq_printf(m, "\n  vm stats threshold: %ld\n",
+					zone->stat_threshold);
 	}
 	return 0;
 }
Index: linux-2.6.18-rc4/include/linux/mmzone.h
===================================================================
--- linux-2.6.18-rc4.orig/include/linux/mmzone.h	2006-08-18 14:52:03.479976166 -0700
+++ linux-2.6.18-rc4/include/linux/mmzone.h	2006-08-18 15:26:11.950215397 -0700
@@ -77,7 +77,6 @@ struct per_cpu_pages {
 struct per_cpu_pageset {
 	struct per_cpu_pages pcp[2];	/* 0: hot.  1: cold */
 #ifdef CONFIG_SMP
-	s8 stat_threshold;	/* maximum diff before update */
 	s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
 #endif
 } ____cacheline_aligned_in_smp;
@@ -154,7 +153,12 @@ struct zone {
 	/*
 	 * zone reclaim becomes active if more unmapped pages exist.
 	 */
-	unsigned long		min_unmapped_ratio;
+	unsigned int		min_unmapped_ratio;
+	/*
+	 * Maximum per processor diff before global update
+	 */
+	unsigned int 		stat_threshold;
+
 	struct per_cpu_pageset	*pageset[NR_CPUS];
 #else
 	struct per_cpu_pageset	pageset[NR_CPUS];