diff options
author | David S. Miller <davem@davemloft.net> | 2008-04-18 02:56:30 -0400 |
---|---|---|
committer | David S. Miller <davem@davemloft.net> | 2008-04-18 02:56:30 -0400 |
commit | 1e42198609d73ed1a9adcba2af275c24c2678420 (patch) | |
tree | 32fd4d9073bfc0f3909af8f9fb4bcff38951d01a /Documentation/scheduler/sched-rt-group.txt | |
parent | 794eb6bf20ebf992c040ea831cd3a9c64b0c1f7a (diff) | |
parent | 4b119e21d0c66c22e8ca03df05d9de623d0eb50f (diff) |
Merge branch 'master' of master.kernel.org:/pub/scm/linux/kernel/git/torvalds/linux-2.6
Diffstat (limited to 'Documentation/scheduler/sched-rt-group.txt')
-rw-r--r-- | Documentation/scheduler/sched-rt-group.txt | 59 |
1 files changed, 59 insertions, 0 deletions
diff --git a/Documentation/scheduler/sched-rt-group.txt b/Documentation/scheduler/sched-rt-group.txt new file mode 100644 index 000000000000..1c6332f4543c --- /dev/null +++ b/Documentation/scheduler/sched-rt-group.txt | |||
@@ -0,0 +1,59 @@ | |||
1 | |||
2 | |||
3 | Real-Time group scheduling. | ||
4 | |||
5 | The problem space: | ||
6 | |||
7 | In order to schedule multiple groups of realtime tasks each group must | ||
8 | be assigned a fixed portion of the CPU time available. Without a minimum | ||
9 | guarantee a realtime group can obviously fall short. A fuzzy upper limit | ||
10 | is of no use since it cannot be relied upon. Which leaves us with just | ||
11 | the single fixed portion. | ||
12 | |||
13 | CPU time is divided by means of specifying how much time can be spent | ||
14 | running in a given period. Say a frame fixed realtime renderer must | ||
15 | deliver 25 frames a second, which yields a period of 0.04s. Now say | ||
16 | it will also have to play some music and respond to input, leaving it | ||
17 | with around 80% for the graphics. We can then give this group a runtime | ||
18 | of 0.8 * 0.04s = 0.032s. | ||
19 | |||
20 | This way the graphics group will have a 0.04s period with a 0.032s runtime | ||
21 | limit. | ||
22 | |||
23 | Now if the audio thread needs to refill the DMA buffer every 0.005s, but | ||
24 | needs only about 3% CPU time to do so, it can do with a 0.03 * 0.005s | ||
25 | = 0.00015s. | ||
26 | |||
27 | |||
28 | The Interface: | ||
29 | |||
30 | system wide: | ||
31 | |||
32 | /proc/sys/kernel/sched_rt_period_ms | ||
33 | /proc/sys/kernel/sched_rt_runtime_us | ||
34 | |||
35 | CONFIG_FAIR_USER_SCHED | ||
36 | |||
37 | /sys/kernel/uids/<uid>/cpu_rt_runtime_us | ||
38 | |||
39 | or | ||
40 | |||
41 | CONFIG_FAIR_CGROUP_SCHED | ||
42 | |||
43 | /cgroup/<cgroup>/cpu.rt_runtime_us | ||
44 | |||
45 | [ time is specified in us because the interface is s32; this gives an | ||
46 | operating range of ~35m to 1us ] | ||
47 | |||
48 | The period takes values in [ 1, INT_MAX ], runtime in [ -1, INT_MAX - 1 ]. | ||
49 | |||
50 | A runtime of -1 specifies runtime == period, ie. no limit. | ||
51 | |||
52 | New groups get the period from /proc/sys/kernel/sched_rt_period_us and | ||
53 | a runtime of 0. | ||
54 | |||
55 | Settings are constrained to: | ||
56 | |||
57 | \Sum_{i} runtime_{i} / global_period <= global_runtime / global_period | ||
58 | |||
59 | in order to keep the configuration schedulable. | ||