[Ksummit-2012-discuss] [ATTEND] memcg and some topics

Kamezawa Hiroyuki kamezawa.hiroyu at jp.fujitsu.com
Tue Jun 19 01:35:01 UTC 2012


I'd like to attend kernel summit to discuss several topics around cgroup/memory cgroup.

- dirty limit, kmem account, soft limit, per-memcg kswapd, etc...

BTW, followings are curious to me in these days but have no idea/patch...

- x86/1GB hugepage.
  How to support this ? SECTION_SIZE should be 1GB ? MAX_ORDER should be ?
  contiguous-memory-allocator ? or using ZONE_MOVABLE ?

- cpu naming/x86

  Now, cpu's ID is determined by the order of MADT. In typical case,
  on 2 thread/ 2 core / 2 socket system..

  cpuid       (thread,core,socket)
  0           (0,0,0)
  1           (0,1,0)
  2           (0,0,1)
  3           (0,1,1)
  4           (1,0,0)
  5           (1,1,0)
  6           (1,0,1)
  7           (1,1,1)

  0-4, 1-5, 2-6, 3-7 shares core.

  HotAdd new cpu socket2 here.

  cpuid
  8           (0,0,2)
  9           (0,1,2)
  10          (1,0,2)
  11          (1,1,2)

  Then, reboot

  0          (0,0,0)
  1          (0,1,0)
  2          (0,0,1)
  3          (0,1,1)
  4          (0,0,2)
  5          (0,1,2)
  .....
cpuid <-> core relation ship will change after reboot. IIUC, there are some
interfaces which uses cpu ID to tell kernel affinity of some components.

So, does anyone have idea for consistent cpu naming or udev-for-cpu-ID ?
For example,
# echo node0.socket1.core1.thread0 > .../cpuset.cpus


Thanks,
-Kame

















More information about the Ksummit-2012-discuss mailing list