Android: automated per-uid task group

Android is a privilege-separated operating system, in which each application runs with a distinct system identity: the Linux user ID (uid).

With this patch (sched-automated-per-uid-task-groups.patch) the kernel automatically creates a distinct task group for each uid (when a process calls set_user()) and places all the tasks that belong to a single uid into the same task group. In this way each application can get a fair amount of the CPU bandwidth (guaranteed by the CFS scheduler), independently by the number of task/threads spawned.

The patch is against the CyanogenMod's 6.1.1 kernel (2.6.35.10) and I tested it successfully on my HTC Desire (Bravo).

I've used the following testcase:

  - run 4 cpu hogs in background as user app_35 (com.android.email in my case):

 # su - app_35

 $ for i in `seq 4`; do yes >/dev/null & done

  - run the Quadrant benchmark in parallel and measure the result with and without the patch applied

Without the patch (output of top):

  PID  PPID USER     STAT   VSZ %MEM CPU %CPU COMMAND

 6533   123 10070    R     202m 48.6   0 20.0 com.aurorasoftworks.quadrant.ui.st

 6506     1 10035    R     1128  0.2   0 20.0 yes

 6507     1 10035    R     1128  0.2   0 20.0 yes

 6508     1 10035    R     1128  0.2   0 20.0 yes

 6509     1 10035    R     1128  0.2   0 20.0 yes

Benchmark result: 676

  uid 10035 (cpu hog)  : 60.0 % cpu quota

 uid 10070 (benchmark): 20.0 % cpu quota

With automated per-uid task group (output of top):

  PID  PPID USER     STAT   VSZ %MEM CPU %CPU COMMAND

 6784   123 10070    S     209m 51.4   0 50.0 com.aurorasoftworks.quadrant.ui.st

 6852     1 10035    R     1128  0.2   0 12.5 yes

 6853     1 10035    R     1128  0.2   0 12.5 yes

 6854     1 10035    R     1128  0.2   0 12.5 yes

 6855     1 10035    R     1128  0.2   0 12.5 yes

Benchmark result: 816

 uid 10035 (cpu hog)  : 50.0 % cpu quota

 uid 10070 (benchmark): 50.0 % cpu quota

Total speedup: ~1.2 (the benchmark is about 20% faster in this case), because in the last case the benchmark gets ~50% of the CPU time and in the other case it gets only ~20%, despite the fact that there are 2 "pair" applications that should be correctly considered as equal from the user's perspective.

This patch is based on the patch "sched: automated per tty task groups" by Mike Galbraith.