1=====================
2CFS Bandwidth Control
3=====================
4
5[ This document only discusses CPU bandwidth control for SCHED_NORMAL.
6  The SCHED_RT case is covered in Documentation/scheduler/sched-rt-group.rst ]
7
8CFS bandwidth control is a CONFIG_FAIR_GROUP_SCHED extension which allows the
9specification of the maximum CPU bandwidth available to a group or hierarchy.
10
11The bandwidth allowed for a group is specified using a quota and period. Within
12each given "period" (microseconds), a group is allowed to consume only up to
13"quota" microseconds of CPU time.  When the CPU bandwidth consumption of a
14group exceeds this limit (for that period), the tasks belonging to its
15hierarchy will be throttled and are not allowed to run again until the next
16period.
17
18A group's unused runtime is globally tracked, being refreshed with quota units
19above at each period boundary.  As threads consume this bandwidth it is
20transferred to cpu-local "silos" on a demand basis.  The amount transferred
21within each of these updates is tunable and described as the "slice".
22
23Management
24----------
25Quota and period are managed within the cpu subsystem via cgroupfs.
26
27cpu.cfs_quota_us: the total available run-time within a period (in microseconds)
28cpu.cfs_period_us: the length of a period (in microseconds)
29cpu.stat: exports throttling statistics [explained further below]
30
31The default values are::
32
33	cpu.cfs_period_us=100ms
34	cpu.cfs_quota=-1
35
36A value of -1 for cpu.cfs_quota_us indicates that the group does not have any
37bandwidth restriction in place, such a group is described as an unconstrained
38bandwidth group.  This represents the traditional work-conserving behavior for
39CFS.
40
41Writing any (valid) positive value(s) will enact the specified bandwidth limit.
42The minimum quota allowed for the quota or period is 1ms.  There is also an
43upper bound on the period length of 1s.  Additional restrictions exist when
44bandwidth limits are used in a hierarchical fashion, these are explained in
45more detail below.
46
47Writing any negative value to cpu.cfs_quota_us will remove the bandwidth limit
48and return the group to an unconstrained state once more.
49
50Any updates to a group's bandwidth specification will result in it becoming
51unthrottled if it is in a constrained state.
52
53System wide settings
54--------------------
55For efficiency run-time is transferred between the global pool and CPU local
56"silos" in a batch fashion.  This greatly reduces global accounting pressure
57on large systems.  The amount transferred each time such an update is required
58is described as the "slice".
59
60This is tunable via procfs::
61
62	/proc/sys/kernel/sched_cfs_bandwidth_slice_us (default=5ms)
63
64Larger slice values will reduce transfer overheads, while smaller values allow
65for more fine-grained consumption.
66
67Statistics
68----------
69A group's bandwidth statistics are exported via 3 fields in cpu.stat.
70
71cpu.stat:
72
73- nr_periods: Number of enforcement intervals that have elapsed.
74- nr_throttled: Number of times the group has been throttled/limited.
75- throttled_time: The total time duration (in nanoseconds) for which entities
76  of the group have been throttled.
77
78This interface is read-only.
79
80Hierarchical considerations
81---------------------------
82The interface enforces that an individual entity's bandwidth is always
83attainable, that is: max(c_i) <= C. However, over-subscription in the
84aggregate case is explicitly allowed to enable work-conserving semantics
85within a hierarchy:
86
87  e.g. \Sum (c_i) may exceed C
88
89[ Where C is the parent's bandwidth, and c_i its children ]
90
91
92There are two ways in which a group may become throttled:
93
94	a. it fully consumes its own quota within a period
95	b. a parent's quota is fully consumed within its period
96
97In case b) above, even though the child may have runtime remaining it will not
98be allowed to until the parent's runtime is refreshed.
99
100Examples
101--------
1021. Limit a group to 1 CPU worth of runtime::
103
104	If period is 250ms and quota is also 250ms, the group will get
105	1 CPU worth of runtime every 250ms.
106
107	# echo 250000 > cpu.cfs_quota_us /* quota = 250ms */
108	# echo 250000 > cpu.cfs_period_us /* period = 250ms */
109
1102. Limit a group to 2 CPUs worth of runtime on a multi-CPU machine
111
112   With 500ms period and 1000ms quota, the group can get 2 CPUs worth of
113   runtime every 500ms::
114
115	# echo 1000000 > cpu.cfs_quota_us /* quota = 1000ms */
116	# echo 500000 > cpu.cfs_period_us /* period = 500ms */
117
118	The larger period here allows for increased burst capacity.
119
1203. Limit a group to 20% of 1 CPU.
121
122   With 50ms period, 10ms quota will be equivalent to 20% of 1 CPU::
123
124	# echo 10000 > cpu.cfs_quota_us /* quota = 10ms */
125	# echo 50000 > cpu.cfs_period_us /* period = 50ms */
126
127   By using a small period here we are ensuring a consistent latency
128   response at the expense of burst capacity.
129