Messages in this thread |  | | From | kamezawa.hiroyu@jp ... | Date | Mon, 16 Jun 2008 18:01:42 +0900 (JST) | Subject | Re: Re: [PATCH 1/6] res_counter: handle limit change |
| |
----- Original Message ----- >> Okay, maye all you want is "don't increase the size of res_counter" > >Actually no, what I want is not to put indirections level when >not required. > "not required" ? I think you miss the point that this patch implements some feedback algorithm in res_counter. If res_counter doesn't support it, Okay, I'll do in memcg. But please see this request from Paul in the prev vers ion. http://marc.info/?l=linux-mm&m=121257010530546&w=2 And what benefits we can get by implementing feedback per subcgroups ?
>But keeping res_counter as small as possible is also my wish. :) > >>>> Is it so strange to add following algorithm in res_counter? >>>> == >>>> set_limit -> fail -> shrink -> set limit -> fail ->shrink >>>> -> success -> return 0 >>>> == >>>> I think this is enough generic. >>> It is, but my point is - we're calling the set_limit (this is a >>> res_counter_resize_limit from your patch, sorry for the confusion again) >>> routine right from the cgroup's write callback and thus can call >>> the desired "ops->shrink_usage" directly, w/o additional level of >>> indirection. >>> >> Hmm, to do that, I'd like to remove strategy function from res_counter. > >Oops... I'm looking at 2.6.26-rc5-mm1's res_counter and don't see such. >I tried to follow the changes in res_counter, but it looks like I've >already missed something. > >What do you mean by "strategy function from res_counter"? > Please ignore. my confusion. "don't call res_counter_write() at set limit" is ok.
Thanks, -Kame
|  |