[RFC][PATCH 0/4[ memcg shrinking usage.

KAMEZAWA Hiroyuki kamezawa.hiroyu at jp.fujitsu.com
Fri Jul 18 03:31:23 PDT 2008


This email is just for dumping my queue before OLS.
(Not testd well. just for sharing idea.)

including following patches. 

[1/4] ... res_counter_set_limit for -EBUSY.
[2/4] ... res_coutner_check_under_val.
[3/4] ... memcg set limit .
[4/4] ... memcg shrink usage. (NEW!)

A brief story for patch 4/4.

 I have been writing background-reclaim support to memcg and used some kernel
 threads. But...
 1. using kernel thread have some complexty and add some amounts of codes.
 2. there are another resource control method, soft-limit is proposed.
    (and will be others.)
 3. wise load balancing between several groups will be hard.

 So, using kernel thread have some problems. Andrew Morton suggested that
 "Do that in user-land. We already have amounts of user-land-helper now."

 Patch 4/4 just adds "shrink_me!" interface. A user land daemon can decrease
 the usage of resource by kicking this. I like this idea. How do you think ?

 Next problem is how to make interaction between userland and the kernel.
 I'm now interested in "inotify". (But have to learn it more ;)

P.S. 
I'll be completely offline until the end of July once I leave Japan.
Please ask/request/blame me directly at Ottawa if you want ;)

See you!
-Kame



More information about the Containers mailing list