Re: [RFC][PATCH 2/4] memcg: use CSS ID in memcg

[KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>]

On Mon, 12 Jan 2009 17:44:24 +0530
Balbir Singh <balbir@linux.vnet.ibm.com> wrote:
					   get_swappiness(next_mem));
> > +	struct mem_cgroup *victim;
> > +	unsigned long start_age;
> > +	int ret, total = 0;
> > +	/*
> > +	 * Reclaim memory from cgroups under root_mem in round robin.
> > +	 */
> > +	start_age = root_mem->scan_age;
> > +
> > +	while (time_after((start_age + 2UL), root_mem->scan_age)) {
> 
> This is confusing, why do we use time_after with scan_age. scan_age
> seems to be incremented every time we scan and has no relationship
> with time. 

time_after() is useful macro for checking counter which can go MAX-1->MAX->0->1>...


> The second thing is what happens if time_after() always
> returns 0, if we've been aggressively scanning? 
That never happens.

> The logic needs some  commenting, why the magic number 2?
> 
memcg->scan_age is update when 
 - the memcg is root of hierarchy.
 - we reclaim memory from memcg.

So, memcg->scan_age is update by 2 means, all memcg under hierarchy is accessed
by reclaim routine.

example) Consider hierarhy like this.
 
          xxx(ID=8)
             /yyy (ID=4)
             /zzz (ID=9)
             /www (ID=3)

In this case, scan will be done in following order

  .....->3->4->8->9->3->4->8->9->...
(start point is determined by last_scanned_child)

everytime we visit "8", 8's scan_age is updated.

So, if we see "8"  2 times, all other groups 4,9,3 is all accessed for freeing
memory. (by me or other threads.)


-Kame

--
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>