Messages in this thread | | | Date | Tue, 6 Mar 2012 17:29:46 -0800 (PST) | From | Hugh Dickins <> | Subject | Re: [PATCH] memcg: fix mapcount check in move charge code for anonymous page |
| |
On Tue, 6 Mar 2012, Naoya Horiguchi wrote: > > IMO, ideally the charge of shared (both file and anon) pages should > be accounted for all cgroups to which the processes mapping the pages > belong to, where each charge is weighted by inverse number of mapcount. > I think accounting total number of mapcount with another counter does > not work, because the weight of charge depends on each page and the > total count of mapcount doesn't describe the proportion among cgroups. > But anyway, it adds more complexity and needs much work, so is not > a short term fix.
That "ideal" complexity was considered before the current memcg approach went in. We elected to go with the less satisfying, but much simpler, single-owner approach, and it does seem to have paid off. I believe that even those who had successfully developed a more complex approach have since abandoned it for performance scalability reasons.
Hugh
| |