| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Date: Fri, 27 Jul 2012 18:20:32 +0800 From: Sha Zhengju <handai.szj@...il.com> To: linux-mm@...ck.org, cgroups@...r.kernel.org Cc: fengguang.wu@...el.com, gthelen@...gle.com, akpm@...ux-foundation.org, yinghan@...gle.com, mhocko@...e.cz, linux-kernel@...r.kernel.org, hannes@...xchg.org, Sha Zhengju <handai.szj@...bao.com> Subject: [PATCH V2 0/6] Per-cgroup page stat accounting From: Sha Zhengju <handai.szj@...bao.com> Hi, list This V2 patch series provide the ability for each memory cgroup to have independent dirty/writeback page statistics which can provide information for per-cgroup direct reclaim or some. In the first three prepare patches, we have done some cleanup and reworked vfs set page dirty routines to make "modify page info" and "dirty page accouting" stay in one function as much as possible for the sake of memcg bigger lock(test numbers are in the specific patch). Kame, I tested these patches on linux mainline v3.5, because I cannot boot up the kernel under linux-next :(. But these patches are cooked on top of your recent memcg patches (I backport them to mainline) and I think there is no hunk with the mm tree. So If there's no other problem, I think it could be considered for merging. Following is performance comparison between before/after the series: Test steps(Mem-24g, ext4): drop_cache; sync cat /proc/meminfo|grep Dirty (=4KB) fio (buffered/randwrite/bs=4k/size=128m/filesize=1g/numjobs=8/sync) cat /proc/meminfo|grep Dirty(=648696kB) We test it for 10 times and get the average numbers: Before: write: io=1024.0MB, bw=334678 KB/s, iops=83669.2 , runt= 3136 msec lat (usec): min=1 , max=26203.1 , avg=81.473, stdev=275.754 After: write: io=1024.0MB, bw=325219 KB/s, iops= 81304.1 , runt= 3226.9 msec lat (usec): min=1 , max=17224 , avg=86.194, stdev=298.183 There is about 2.8% performance decrease. But I notice that once memcg is enabled, the root_memcg exsits and all pages allocated are belong to it, so they will go through the root memcg statistics routines which bring some overhead. Moreover,in case of memcg_is_enable && no cgroups, we can get root memcg stats just from global numbers which can avoid both accounting overheads and many if-test overheads. Later I'll work further into it. Any comments are welcomed. : ) Change log: v2 <-- v1: 1. add test numbers 2. some small fix and comments Sha Zhengju (6): memcg-remove-MEMCG_NR_FILE_MAPPED.patch Make-TestSetPageDirty-and-dirty-page-accounting-in-o.patch Use-vfs-__set_page_dirty-interface-instead-of-doing-.patch memcg-add-per-cgroup-dirty-pages-accounting.patch memcg-add-per-cgroup-writeback-pages-accounting.patch memcg-Document-cgroup-dirty-writeback-memory-statist.patch Documentation/cgroups/memory.txt | 2 + fs/buffer.c | 36 +++++++++++++++-------- fs/ceph/addr.c | 20 +------------ include/linux/buffer_head.h | 2 + include/linux/memcontrol.h | 30 ++++++++++++++----- mm/filemap.c | 9 ++++++ mm/memcontrol.c | 58 +++++++++++++++++++------------------- mm/page-writeback.c | 48 ++++++++++++++++++++++++++++--- mm/rmap.c | 4 +- mm/truncate.c | 6 ++++ 10 files changed, 141 insertions(+), 74 deletions(-) -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@...r.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Powered by blists - more mailing lists