6 This document only discusses CPU bandwidth control for SCHED_NORMAL.
7 The SCHED_RT case is covered in Documentation/scheduler/sched-rt-group.rst
9 CFS bandwidth control is a CONFIG_FAIR_GROUP_SCHED extension which allows the
10 specification of the maximum CPU bandwidth available to a group or hierarchy.
12 The bandwidth allowed for a group is specified using a quota and period. Within
13 each given "period" (microseconds), a task group is allocated up to "quota"
14 microseconds of CPU time. That quota is assigned to per-cpu run queues in
15 slices as threads in the cgroup become runnable. Once all quota has been
16 assigned any additional requests for quota will result in those threads being
17 throttled. Throttled threads will not be able to run again until the next
18 period when the quota is replenished.
20 A group's unassigned quota is globally tracked, being refreshed back to
21 cfs_quota units at each period boundary. As threads consume this bandwidth it
22 is transferred to cpu-local "silos" on a demand basis. The amount transferred
23 within each of these updates is tunable and described as the "slice".
27 Quota and period are managed within the cpu subsystem via cgroupfs.
30 The cgroupfs files described in this section are only applicable
31 to cgroup v1. For cgroup v2, see
32 :ref:`Documentation/admin-guide/cgroup-v2.rst <cgroup-v2-cpu>`.
34 - cpu.cfs_quota_us: the total available run-time within a period (in
36 - cpu.cfs_period_us: the length of a period (in microseconds)
37 - cpu.stat: exports throttling statistics [explained further below]
39 The default values are::
41 cpu.cfs_period_us=100ms
44 A value of -1 for cpu.cfs_quota_us indicates that the group does not have any
45 bandwidth restriction in place, such a group is described as an unconstrained
46 bandwidth group. This represents the traditional work-conserving behavior for
49 Writing any (valid) positive value(s) will enact the specified bandwidth limit.
50 The minimum quota allowed for the quota or period is 1ms. There is also an
51 upper bound on the period length of 1s. Additional restrictions exist when
52 bandwidth limits are used in a hierarchical fashion, these are explained in
55 Writing any negative value to cpu.cfs_quota_us will remove the bandwidth limit
56 and return the group to an unconstrained state once more.
58 Any updates to a group's bandwidth specification will result in it becoming
59 unthrottled if it is in a constrained state.
63 For efficiency run-time is transferred between the global pool and CPU local
64 "silos" in a batch fashion. This greatly reduces global accounting pressure
65 on large systems. The amount transferred each time such an update is required
66 is described as the "slice".
68 This is tunable via procfs::
70 /proc/sys/kernel/sched_cfs_bandwidth_slice_us (default=5ms)
72 Larger slice values will reduce transfer overheads, while smaller values allow
73 for more fine-grained consumption.
77 A group's bandwidth statistics are exported via 3 fields in cpu.stat.
81 - nr_periods: Number of enforcement intervals that have elapsed.
82 - nr_throttled: Number of times the group has been throttled/limited.
83 - throttled_time: The total time duration (in nanoseconds) for which entities
84 of the group have been throttled.
86 This interface is read-only.
88 Hierarchical considerations
89 ---------------------------
90 The interface enforces that an individual entity's bandwidth is always
91 attainable, that is: max(c_i) <= C. However, over-subscription in the
92 aggregate case is explicitly allowed to enable work-conserving semantics
95 e.g. \Sum (c_i) may exceed C
97 [ Where C is the parent's bandwidth, and c_i its children ]
100 There are two ways in which a group may become throttled:
102 a. it fully consumes its own quota within a period
103 b. a parent's quota is fully consumed within its period
105 In case b) above, even though the child may have runtime remaining it will not
106 be allowed to until the parent's runtime is refreshed.
108 CFS Bandwidth Quota Caveats
109 ---------------------------
110 Once a slice is assigned to a cpu it does not expire. However all but 1ms of
111 the slice may be returned to the global pool if all threads on that cpu become
112 unrunnable. This is configured at compile time by the min_cfs_rq_runtime
113 variable. This is a performance tweak that helps prevent added contention on
116 The fact that cpu-local slices do not expire results in some interesting corner
117 cases that should be understood.
119 For cgroup cpu constrained applications that are cpu limited this is a
120 relatively moot point because they will naturally consume the entirety of their
121 quota as well as the entirety of each cpu-local slice in each period. As a
122 result it is expected that nr_periods roughly equal nr_throttled, and that
123 cpuacct.usage will increase roughly equal to cfs_quota_us in each period.
125 For highly-threaded, non-cpu bound applications this non-expiration nuance
126 allows applications to briefly burst past their quota limits by the amount of
127 unused slice on each cpu that the task group is running on (typically at most
128 1ms per cpu or as defined by min_cfs_rq_runtime). This slight burst only
129 applies if quota had been assigned to a cpu and then not fully used or returned
130 in previous periods. This burst amount will not be transferred between cores.
131 As a result, this mechanism still strictly limits the task group to quota
132 average usage, albeit over a longer time window than a single period. This
133 also limits the burst ability to no more than 1ms per cpu. This provides
134 better more predictable user experience for highly threaded applications with
135 small quota limits on high core count machines. It also eliminates the
136 propensity to throttle these applications while simultanously using less than
137 quota amounts of cpu. Another way to say this, is that by allowing the unused
138 portion of a slice to remain valid across periods we have decreased the
139 possibility of wastefully expiring quota on cpu-local silos that don't need a
140 full slice's amount of cpu time.
142 The interaction between cpu-bound and non-cpu-bound-interactive applications
143 should also be considered, especially when single core usage hits 100%. If you
144 gave each of these applications half of a cpu-core and they both got scheduled
145 on the same CPU it is theoretically possible that the non-cpu bound application
146 will use up to 1ms additional quota in some periods, thereby preventing the
147 cpu-bound application from fully using its quota by that same amount. In these
148 instances it will be up to the CFS algorithm (see sched-design-CFS.rst) to
149 decide which application is chosen to run, as they will both be runnable and
150 have remaining quota. This runtime discrepancy will be made up in the following
151 periods when the interactive application idles.
155 1. Limit a group to 1 CPU worth of runtime::
157 If period is 250ms and quota is also 250ms, the group will get
158 1 CPU worth of runtime every 250ms.
160 # echo 250000 > cpu.cfs_quota_us /* quota = 250ms */
161 # echo 250000 > cpu.cfs_period_us /* period = 250ms */
163 2. Limit a group to 2 CPUs worth of runtime on a multi-CPU machine
165 With 500ms period and 1000ms quota, the group can get 2 CPUs worth of
166 runtime every 500ms::
168 # echo 1000000 > cpu.cfs_quota_us /* quota = 1000ms */
169 # echo 500000 > cpu.cfs_period_us /* period = 500ms */
171 The larger period here allows for increased burst capacity.
173 3. Limit a group to 20% of 1 CPU.
175 With 50ms period, 10ms quota will be equivalent to 20% of 1 CPU::
177 # echo 10000 > cpu.cfs_quota_us /* quota = 10ms */
178 # echo 50000 > cpu.cfs_period_us /* period = 50ms */
180 By using a small period here we are ensuring a consistent latency
181 response at the expense of burst capacity.