Messages in this thread |  | | | Date | Wed, 17 Sep 2008 18:40:08 -0700 | | From | Andrew Morton <> | | Subject | Re: [RFC][PATCH] Remove cgroup member from struct page (v3) |
| |
On Wed, 17 Sep 2008 16:28:26 -0700 Balbir Singh <balbir@linux.vnet.ibm.com> wrote:
> > Before trying the sparsemem approach, I tried a radix tree per node, > per zone and I seemed to actually get some performance > improvement.(1.5% (noise maybe)) > > But please do see and review (tested on my x86_64 box with unixbench > and some other simple tests) > > v4..v3 > 1. Use a radix tree per node, per zone > > v3...v2 > 1. Convert flags to unsigned long > 2. Move page_cgroup->lock to a bit spin lock in flags > > v2...v1 > > 1. Fix a small bug, don't call radix_tree_preload_end(), if preload fails > > This is a rewrite of a patch I had written long back to remove struct page > (I shared the patches with Kamezawa, but never posted them anywhere else). > I spent the weekend, cleaning them up for 2.6.27-rc5-mmotm (29 Aug 2008). > > I've tested the patches on an x86_64 box, I've run a simple test running > under the memory control group and the same test running concurrently under > two different groups (and creating pressure within their groups). > > Advantages of the patch > > 1. It removes the extra pointer in struct page > > Disadvantages > > 1. Radix tree lookup is not an O(1) operation, once the page is known > getting to the page_cgroup (pc) is a little more expensive now.
Why are we doing this? I can guess, but I'd rather not have to.
a) It's slower.
b) It uses even more memory worst-case.
c) It uses less memory best-case.
someone somewhere decided that (Aa + Bb) / Cc < 1.0. What are the values of A, B and C and where did they come from? ;)
(IOW, your changelog is in the category "sucky", along with 90% of the others)
|  |