1 ===================== 1 =====================
2 CFS Bandwidth Control 2 CFS Bandwidth Control
3 ===================== 3 =====================
4 4
5 .. note:: 5 .. note::
6 This document only discusses CPU bandwidth 6 This document only discusses CPU bandwidth control for SCHED_NORMAL.
7 The SCHED_RT case is covered in Documentati 7 The SCHED_RT case is covered in Documentation/scheduler/sched-rt-group.rst
8 8
9 CFS bandwidth control is a CONFIG_FAIR_GROUP_S 9 CFS bandwidth control is a CONFIG_FAIR_GROUP_SCHED extension which allows the
10 specification of the maximum CPU bandwidth ava 10 specification of the maximum CPU bandwidth available to a group or hierarchy.
11 11
12 The bandwidth allowed for a group is specified 12 The bandwidth allowed for a group is specified using a quota and period. Within
13 each given "period" (microseconds), a task gro 13 each given "period" (microseconds), a task group is allocated up to "quota"
14 microseconds of CPU time. That quota is assign 14 microseconds of CPU time. That quota is assigned to per-cpu run queues in
15 slices as threads in the cgroup become runnabl 15 slices as threads in the cgroup become runnable. Once all quota has been
16 assigned any additional requests for quota wil 16 assigned any additional requests for quota will result in those threads being
17 throttled. Throttled threads will not be able 17 throttled. Throttled threads will not be able to run again until the next
18 period when the quota is replenished. 18 period when the quota is replenished.
19 19
20 A group's unassigned quota is globally tracked 20 A group's unassigned quota is globally tracked, being refreshed back to
21 cfs_quota units at each period boundary. As th 21 cfs_quota units at each period boundary. As threads consume this bandwidth it
22 is transferred to cpu-local "silos" on a deman 22 is transferred to cpu-local "silos" on a demand basis. The amount transferred
23 within each of these updates is tunable and de 23 within each of these updates is tunable and described as the "slice".
24 24
25 Burst feature <<
26 ------------- <<
27 This feature borrows time now against our futu <<
28 increased interference against the other syste <<
29 <<
30 Traditional (UP-EDF) bandwidth control is some <<
31 <<
32 (U = \Sum u_i) <= 1 <<
33 <<
34 This guaranteeds both that every deadline is m <<
35 stable. After all, if U were > 1, then for eve <<
36 we'd have to run more than a second of program <<
37 our deadline, but the next deadline will be fu <<
38 never time to catch up, unbounded fail. <<
39 <<
40 The burst feature observes that a workload doe <<
41 quota; this enables one to describe u_i as a s <<
42 <<
43 For example, have u_i = {x,e}_i, where x is th <<
44 (the traditional WCET). This effectively allow <<
45 increasing the efficiency (we can pack more ta <<
46 the cost of missing deadlines when all the odd <<
47 does maintain stability, since every overrun m <<
48 underrun as long as our x is above the average <<
49 <<
50 That is, suppose we have 2 tasks, both specify <<
51 have a p(95)*p(95) = 90.25% chance both tasks <<
52 everything is good. At the same time we have a <<
53 both tasks will exceed their quota at the same <<
54 fail). Somewhere in between there's a threshol <<
55 the other doesn't underrun enough to compensat <<
56 specific CDFs. <<
57 <<
58 At the same time, we can say that the worst ca <<
59 \Sum e_i; that is, there is a bounded tardines <<
60 that x+e is indeed WCET). <<
61 <<
62 The interferenece when using burst is valued b <<
63 missing the deadline and the average WCET. Tes <<
64 there many cgroups or CPU is under utilized, t <<
65 limited. More details are shown in: <<
66 https://lore.kernel.org/lkml/5371BD36-55AE-4F7 <<
67 <<
68 Management 25 Management
69 ---------- 26 ----------
70 Quota, period and burst are managed within the !! 27 Quota and period are managed within the cpu subsystem via cgroupfs.
71 28
72 .. note:: 29 .. note::
73 The cgroupfs files described in this sectio 30 The cgroupfs files described in this section are only applicable
74 to cgroup v1. For cgroup v2, see 31 to cgroup v1. For cgroup v2, see
75 :ref:`Documentation/admin-guide/cgroup-v2.r 32 :ref:`Documentation/admin-guide/cgroup-v2.rst <cgroup-v2-cpu>`.
76 33
77 - cpu.cfs_quota_us: run-time replenished withi !! 34 - cpu.cfs_quota_us: the total available run-time within a period (in
>> 35 microseconds)
78 - cpu.cfs_period_us: the length of a period (i 36 - cpu.cfs_period_us: the length of a period (in microseconds)
79 - cpu.stat: exports throttling statistics [exp 37 - cpu.stat: exports throttling statistics [explained further below]
80 - cpu.cfs_burst_us: the maximum accumulated ru <<
81 38
82 The default values are:: 39 The default values are::
83 40
84 cpu.cfs_period_us=100ms 41 cpu.cfs_period_us=100ms
85 cpu.cfs_quota_us=-1 !! 42 cpu.cfs_quota=-1
86 cpu.cfs_burst_us=0 <<
87 43
88 A value of -1 for cpu.cfs_quota_us indicates t 44 A value of -1 for cpu.cfs_quota_us indicates that the group does not have any
89 bandwidth restriction in place, such a group i 45 bandwidth restriction in place, such a group is described as an unconstrained
90 bandwidth group. This represents the tradition 46 bandwidth group. This represents the traditional work-conserving behavior for
91 CFS. 47 CFS.
92 48
93 Writing any (valid) positive value(s) no small !! 49 Writing any (valid) positive value(s) will enact the specified bandwidth limit.
94 enact the specified bandwidth limit. The minim !! 50 The minimum quota allowed for the quota or period is 1ms. There is also an
95 period is 1ms. There is also an upper bound on !! 51 upper bound on the period length of 1s. Additional restrictions exist when
96 Additional restrictions exist when bandwidth l !! 52 bandwidth limits are used in a hierarchical fashion, these are explained in
97 fashion, these are explained in more detail be !! 53 more detail below.
98 54
99 Writing any negative value to cpu.cfs_quota_us 55 Writing any negative value to cpu.cfs_quota_us will remove the bandwidth limit
100 and return the group to an unconstrained state 56 and return the group to an unconstrained state once more.
101 57
102 A value of 0 for cpu.cfs_burst_us indicates th <<
103 any unused bandwidth. It makes the traditional <<
104 CFS unchanged. Writing any (valid) positive va <<
105 cpu.cfs_quota_us into cpu.cfs_burst_us will en <<
106 accumulation. <<
107 <<
108 Any updates to a group's bandwidth specificati 58 Any updates to a group's bandwidth specification will result in it becoming
109 unthrottled if it is in a constrained state. 59 unthrottled if it is in a constrained state.
110 60
111 System wide settings 61 System wide settings
112 -------------------- 62 --------------------
113 For efficiency run-time is transferred between 63 For efficiency run-time is transferred between the global pool and CPU local
114 "silos" in a batch fashion. This greatly reduc 64 "silos" in a batch fashion. This greatly reduces global accounting pressure
115 on large systems. The amount transferred each 65 on large systems. The amount transferred each time such an update is required
116 is described as the "slice". 66 is described as the "slice".
117 67
118 This is tunable via procfs:: 68 This is tunable via procfs::
119 69
120 /proc/sys/kernel/sched_cfs_bandwidth_s 70 /proc/sys/kernel/sched_cfs_bandwidth_slice_us (default=5ms)
121 71
122 Larger slice values will reduce transfer overh 72 Larger slice values will reduce transfer overheads, while smaller values allow
123 for more fine-grained consumption. 73 for more fine-grained consumption.
124 74
125 Statistics 75 Statistics
126 ---------- 76 ----------
127 A group's bandwidth statistics are exported vi !! 77 A group's bandwidth statistics are exported via 3 fields in cpu.stat.
128 78
129 cpu.stat: 79 cpu.stat:
130 80
131 - nr_periods: Number of enforcement intervals 81 - nr_periods: Number of enforcement intervals that have elapsed.
132 - nr_throttled: Number of times the group has 82 - nr_throttled: Number of times the group has been throttled/limited.
133 - throttled_time: The total time duration (in 83 - throttled_time: The total time duration (in nanoseconds) for which entities
134 of the group have been throttled. 84 of the group have been throttled.
135 - nr_bursts: Number of periods burst occurs. <<
136 - burst_time: Cumulative wall-time (in nanosec <<
137 above quota in respective periods. <<
138 85
139 This interface is read-only. 86 This interface is read-only.
140 87
141 Hierarchical considerations 88 Hierarchical considerations
142 --------------------------- 89 ---------------------------
143 The interface enforces that an individual enti 90 The interface enforces that an individual entity's bandwidth is always
144 attainable, that is: max(c_i) <= C. However, o 91 attainable, that is: max(c_i) <= C. However, over-subscription in the
145 aggregate case is explicitly allowed to enable 92 aggregate case is explicitly allowed to enable work-conserving semantics
146 within a hierarchy: 93 within a hierarchy:
147 94
148 e.g. \Sum (c_i) may exceed C 95 e.g. \Sum (c_i) may exceed C
149 96
150 [ Where C is the parent's bandwidth, and c_i i 97 [ Where C is the parent's bandwidth, and c_i its children ]
151 98
152 99
153 There are two ways in which a group may become 100 There are two ways in which a group may become throttled:
154 101
155 a. it fully consumes its own quota wit 102 a. it fully consumes its own quota within a period
156 b. a parent's quota is fully consumed 103 b. a parent's quota is fully consumed within its period
157 104
158 In case b) above, even though the child may ha 105 In case b) above, even though the child may have runtime remaining it will not
159 be allowed to until the parent's runtime is re 106 be allowed to until the parent's runtime is refreshed.
160 107
161 CFS Bandwidth Quota Caveats 108 CFS Bandwidth Quota Caveats
162 --------------------------- 109 ---------------------------
163 Once a slice is assigned to a cpu it does not 110 Once a slice is assigned to a cpu it does not expire. However all but 1ms of
164 the slice may be returned to the global pool i 111 the slice may be returned to the global pool if all threads on that cpu become
165 unrunnable. This is configured at compile time 112 unrunnable. This is configured at compile time by the min_cfs_rq_runtime
166 variable. This is a performance tweak that hel 113 variable. This is a performance tweak that helps prevent added contention on
167 the global lock. 114 the global lock.
168 115
169 The fact that cpu-local slices do not expire r 116 The fact that cpu-local slices do not expire results in some interesting corner
170 cases that should be understood. 117 cases that should be understood.
171 118
172 For cgroup cpu constrained applications that a 119 For cgroup cpu constrained applications that are cpu limited this is a
173 relatively moot point because they will natura 120 relatively moot point because they will naturally consume the entirety of their
174 quota as well as the entirety of each cpu-loca 121 quota as well as the entirety of each cpu-local slice in each period. As a
175 result it is expected that nr_periods roughly 122 result it is expected that nr_periods roughly equal nr_throttled, and that
176 cpuacct.usage will increase roughly equal to c 123 cpuacct.usage will increase roughly equal to cfs_quota_us in each period.
177 124
178 For highly-threaded, non-cpu bound application 125 For highly-threaded, non-cpu bound applications this non-expiration nuance
179 allows applications to briefly burst past thei 126 allows applications to briefly burst past their quota limits by the amount of
180 unused slice on each cpu that the task group i 127 unused slice on each cpu that the task group is running on (typically at most
181 1ms per cpu or as defined by min_cfs_rq_runtim 128 1ms per cpu or as defined by min_cfs_rq_runtime). This slight burst only
182 applies if quota had been assigned to a cpu an 129 applies if quota had been assigned to a cpu and then not fully used or returned
183 in previous periods. This burst amount will no 130 in previous periods. This burst amount will not be transferred between cores.
184 As a result, this mechanism still strictly lim 131 As a result, this mechanism still strictly limits the task group to quota
185 average usage, albeit over a longer time windo 132 average usage, albeit over a longer time window than a single period. This
186 also limits the burst ability to no more than 133 also limits the burst ability to no more than 1ms per cpu. This provides
187 better more predictable user experience for hi 134 better more predictable user experience for highly threaded applications with
188 small quota limits on high core count machines 135 small quota limits on high core count machines. It also eliminates the
189 propensity to throttle these applications whil !! 136 propensity to throttle these applications while simultanously using less than
190 quota amounts of cpu. Another way to say this, 137 quota amounts of cpu. Another way to say this, is that by allowing the unused
191 portion of a slice to remain valid across peri 138 portion of a slice to remain valid across periods we have decreased the
192 possibility of wastefully expiring quota on cp 139 possibility of wastefully expiring quota on cpu-local silos that don't need a
193 full slice's amount of cpu time. 140 full slice's amount of cpu time.
194 141
195 The interaction between cpu-bound and non-cpu- 142 The interaction between cpu-bound and non-cpu-bound-interactive applications
196 should also be considered, especially when sin 143 should also be considered, especially when single core usage hits 100%. If you
197 gave each of these applications half of a cpu- 144 gave each of these applications half of a cpu-core and they both got scheduled
198 on the same CPU it is theoretically possible t 145 on the same CPU it is theoretically possible that the non-cpu bound application
199 will use up to 1ms additional quota in some pe 146 will use up to 1ms additional quota in some periods, thereby preventing the
200 cpu-bound application from fully using its quo 147 cpu-bound application from fully using its quota by that same amount. In these
201 instances it will be up to the CFS algorithm ( 148 instances it will be up to the CFS algorithm (see sched-design-CFS.rst) to
202 decide which application is chosen to run, as 149 decide which application is chosen to run, as they will both be runnable and
203 have remaining quota. This runtime discrepancy 150 have remaining quota. This runtime discrepancy will be made up in the following
204 periods when the interactive application idles 151 periods when the interactive application idles.
205 152
206 Examples 153 Examples
207 -------- 154 --------
208 1. Limit a group to 1 CPU worth of runtime:: 155 1. Limit a group to 1 CPU worth of runtime::
209 156
210 If period is 250ms and quota is also 2 157 If period is 250ms and quota is also 250ms, the group will get
211 1 CPU worth of runtime every 250ms. 158 1 CPU worth of runtime every 250ms.
212 159
213 # echo 250000 > cpu.cfs_quota_us /* qu 160 # echo 250000 > cpu.cfs_quota_us /* quota = 250ms */
214 # echo 250000 > cpu.cfs_period_us /* p 161 # echo 250000 > cpu.cfs_period_us /* period = 250ms */
215 162
216 2. Limit a group to 2 CPUs worth of runtime on 163 2. Limit a group to 2 CPUs worth of runtime on a multi-CPU machine
217 164
218 With 500ms period and 1000ms quota, the gro 165 With 500ms period and 1000ms quota, the group can get 2 CPUs worth of
219 runtime every 500ms:: 166 runtime every 500ms::
220 167
221 # echo 1000000 > cpu.cfs_quota_us /* q 168 # echo 1000000 > cpu.cfs_quota_us /* quota = 1000ms */
222 # echo 500000 > cpu.cfs_period_us /* p 169 # echo 500000 > cpu.cfs_period_us /* period = 500ms */
223 170
224 The larger period here allows for incr 171 The larger period here allows for increased burst capacity.
225 172
226 3. Limit a group to 20% of 1 CPU. 173 3. Limit a group to 20% of 1 CPU.
227 174
228 With 50ms period, 10ms quota will be equiva 175 With 50ms period, 10ms quota will be equivalent to 20% of 1 CPU::
229 176
230 # echo 10000 > cpu.cfs_quota_us /* quo 177 # echo 10000 > cpu.cfs_quota_us /* quota = 10ms */
231 # echo 50000 > cpu.cfs_period_us /* pe 178 # echo 50000 > cpu.cfs_period_us /* period = 50ms */
232 179
233 By using a small period here we are ensurin 180 By using a small period here we are ensuring a consistent latency
234 response at the expense of burst capacity. 181 response at the expense of burst capacity.
235 <<
236 4. Limit a group to 40% of 1 CPU, and allow ac <<
237 additionally, in case accumulation has been <<
238 <<
239 With 50ms period, 20ms quota will be equiva <<
240 And 10ms burst will be equivalent to 20% of <<
241 <<
242 # echo 20000 > cpu.cfs_quota_us /* quo <<
243 # echo 50000 > cpu.cfs_period_us /* pe <<
244 # echo 10000 > cpu.cfs_burst_us /* bur <<
245 <<
246 Larger buffer setting (no larger than quota <<
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.