Thread (34 messages) 34 messages, 4 authors, 2011-06-27

Re: [PATCH 4/7] memcg: update numa information based on event counter

From: Michal Hocko <hidden>
Date: 2011-06-23 08:12:12
Also in: lkml

On Thu 23-06-11 15:27:34, KAMEZAWA Hiroyuki wrote:
On Wed, 22 Jun 2011 17:53:09 +0200
Michal Hocko [off-list ref] wrote:
quoted
On Thu 16-06-11 12:54:00, KAMEZAWA Hiroyuki wrote:
[...]
quoted
quoted
@@ -1616,17 +1629,32 @@ mem_cgroup_select_victim(struct mem_cgro
 /*
  * Always updating the nodemask is not very good - even if we have an empty
  * list or the wrong list here, we can start from some node and traverse all
- * nodes based on the zonelist. So update the list loosely once per 10 secs.
+ * nodes based on the zonelist.
  *
+ * The counter of mem->numascan_update is updated once per
+ * NUMASCAN_EVENTS_TARGET. We update the numa information when we see
+ * the number of event is larger than 3% of limit or  64MB pagein/pageout.
  */
+#define NUMASCAN_UPDATE_RATIO	(3)
+#define NUMASCAN_UPDATE_THRESH	(16384UL) /* 16k events of pagein/pageout */
 static void mem_cgroup_may_update_nodemask(struct mem_cgroup *mem)
 {
 	int nid;
-
-	if (time_after(mem->next_scan_node_update, jiffies))
+	unsigned long long limit;
+	/* if no limit, we never reach here */
+	limit = res_counter_read_u64(&mem->res, RES_LIMIT);
+	limit /= PAGE_SIZE;
+	/* 3% of limit */
+	limit = (limit * NUMASCAN_UPDATE_RATIO/100UL);
+	limit = min_t(unsigned long long, limit, NUMASCAN_UPDATE_THRESH);
+	/*
+	 * If the number of pagein/out event is larger than 3% of limit or
+	 * 64MB pagein/out, refresh numa information.
+	 */
+	if (atomic_read(&mem->numascan_update) < limit ||
+	    !mutex_trylock(&mem->numascan_mutex))
 		return;
I am not sure whether a mutex is not overkill here. What about using an
atomic operation instead?
I think mutex is informative than atomic counter for code readers.
If influence of overhead is not big, I'd like to use mutex.
I do not have a strong opinion on that. mem_cgroup is not that
widespread structure to think about every single byte. On the other hand
atomic test&set would do the same thing. We are already using atomic
operations to manipulate numascan_update so doing it whole atomic based
sounds natural.

-- 
Michal Hocko
SUSE Labs
SUSE LINUX s.r.o.
Lihovarska 1060/12
190 00 Praha 9    
Czech Republic

--
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/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help