On Tue, 18 Jan 2011 12:02:51 -0800 Ying Han <yinghan@xxxxxxxxxx> wrote: > On Thu, Jan 13, 2011 at 4:11 PM, KAMEZAWA Hiroyuki > <kamezawa.hiroyu@xxxxxxxxxxxxxx> wrote: > > > > Please explain your handling of 'hierarchy' in description. > I haven't thought through the 'hierarchy' handling in this patchset > which I will probably put more thoughts in the following > posts. Do you have recommendations on handing the 'hierarchy' ? > For example, assume a Hierarchy like following. A \ B B's usage is accoutned into A, too. So, it's difficult to determine when A's kswapd should run if - A's kswapd runs only against 'A' - A's kswapd just see information of A's LRU - B has its own kswapd...this means A has 2 kswapd. ..... What I think are 2 options. (1) having one kswapd per hierarchy, IOW, B will never have hierarchy. or (2) having kswapd per cgroup but it shares mutex. Parent's kswapd will never run if one of children's run. (1) sounds slow and handling of children's watermark will be serialized. (2) sounds we may have too much worker. I like something between (1) and (2) ;) sqrt(num_of_cgroup) of kswapd is good ? Thanks, -Kame -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxxx For more info on Linux MM, see: http://www.linux-mm.org/ . Fight unfair telecom policy in Canada: sign http://dissolvethecrtc.ca/ Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>