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 25 Burst feature
26 ------------- 26 -------------
27 This feature borrows time now against our futu 27 This feature borrows time now against our future underrun, at the cost of
28 increased interference against the other syste 28 increased interference against the other system users. All nicely bounded.
29 29
30 Traditional (UP-EDF) bandwidth control is some 30 Traditional (UP-EDF) bandwidth control is something like:
31 31
32 (U = \Sum u_i) <= 1 32 (U = \Sum u_i) <= 1
33 33
34 This guaranteeds both that every deadline is m 34 This guaranteeds both that every deadline is met and that the system is
35 stable. After all, if U were > 1, then for eve 35 stable. After all, if U were > 1, then for every second of walltime,
36 we'd have to run more than a second of program 36 we'd have to run more than a second of program time, and obviously miss
37 our deadline, but the next deadline will be fu 37 our deadline, but the next deadline will be further out still, there is
38 never time to catch up, unbounded fail. 38 never time to catch up, unbounded fail.
39 39
40 The burst feature observes that a workload doe 40 The burst feature observes that a workload doesn't always executes the full
41 quota; this enables one to describe u_i as a s 41 quota; this enables one to describe u_i as a statistical distribution.
42 42
43 For example, have u_i = {x,e}_i, where x is th 43 For example, have u_i = {x,e}_i, where x is the p(95) and x+e p(100)
44 (the traditional WCET). This effectively allow 44 (the traditional WCET). This effectively allows u to be smaller,
45 increasing the efficiency (we can pack more ta 45 increasing the efficiency (we can pack more tasks in the system), but at
46 the cost of missing deadlines when all the odd 46 the cost of missing deadlines when all the odds line up. However, it
47 does maintain stability, since every overrun m 47 does maintain stability, since every overrun must be paired with an
48 underrun as long as our x is above the average 48 underrun as long as our x is above the average.
49 49
50 That is, suppose we have 2 tasks, both specify 50 That is, suppose we have 2 tasks, both specify a p(95) value, then we
51 have a p(95)*p(95) = 90.25% chance both tasks 51 have a p(95)*p(95) = 90.25% chance both tasks are within their quota and
52 everything is good. At the same time we have a 52 everything is good. At the same time we have a p(5)p(5) = 0.25% chance
53 both tasks will exceed their quota at the same 53 both tasks will exceed their quota at the same time (guaranteed deadline
54 fail). Somewhere in between there's a threshol 54 fail). Somewhere in between there's a threshold where one exceeds and
55 the other doesn't underrun enough to compensat 55 the other doesn't underrun enough to compensate; this depends on the
56 specific CDFs. 56 specific CDFs.
57 57
58 At the same time, we can say that the worst ca 58 At the same time, we can say that the worst case deadline miss, will be
59 \Sum e_i; that is, there is a bounded tardines 59 \Sum e_i; that is, there is a bounded tardiness (under the assumption
60 that x+e is indeed WCET). 60 that x+e is indeed WCET).
61 61
62 The interferenece when using burst is valued b 62 The interferenece when using burst is valued by the possibilities for
63 missing the deadline and the average WCET. Tes 63 missing the deadline and the average WCET. Test results showed that when
64 there many cgroups or CPU is under utilized, t 64 there many cgroups or CPU is under utilized, the interference is
65 limited. More details are shown in: 65 limited. More details are shown in:
66 https://lore.kernel.org/lkml/5371BD36-55AE-4F7 66 https://lore.kernel.org/lkml/5371BD36-55AE-4F71-B9D7-B86DC32E3D2B@linux.alibaba.com/
67 67
68 Management 68 Management
69 ---------- 69 ----------
70 Quota, period and burst are managed within the 70 Quota, period and burst are managed within the cpu subsystem via cgroupfs.
71 71
72 .. note:: 72 .. note::
73 The cgroupfs files described in this sectio 73 The cgroupfs files described in this section are only applicable
74 to cgroup v1. For cgroup v2, see 74 to cgroup v1. For cgroup v2, see
75 :ref:`Documentation/admin-guide/cgroup-v2.r 75 :ref:`Documentation/admin-guide/cgroup-v2.rst <cgroup-v2-cpu>`.
76 76
77 - cpu.cfs_quota_us: run-time replenished withi 77 - cpu.cfs_quota_us: run-time replenished within a period (in microseconds)
78 - cpu.cfs_period_us: the length of a period (i 78 - cpu.cfs_period_us: the length of a period (in microseconds)
79 - cpu.stat: exports throttling statistics [exp 79 - cpu.stat: exports throttling statistics [explained further below]
80 - cpu.cfs_burst_us: the maximum accumulated ru 80 - cpu.cfs_burst_us: the maximum accumulated run-time (in microseconds)
81 81
82 The default values are:: 82 The default values are::
83 83
84 cpu.cfs_period_us=100ms 84 cpu.cfs_period_us=100ms
85 cpu.cfs_quota_us=-1 85 cpu.cfs_quota_us=-1
86 cpu.cfs_burst_us=0 86 cpu.cfs_burst_us=0
87 87
88 A value of -1 for cpu.cfs_quota_us indicates t 88 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 89 bandwidth restriction in place, such a group is described as an unconstrained
90 bandwidth group. This represents the tradition 90 bandwidth group. This represents the traditional work-conserving behavior for
91 CFS. 91 CFS.
92 92
93 Writing any (valid) positive value(s) no small 93 Writing any (valid) positive value(s) no smaller than cpu.cfs_burst_us will
94 enact the specified bandwidth limit. The minim 94 enact the specified bandwidth limit. The minimum quota allowed for the quota or
95 period is 1ms. There is also an upper bound on 95 period is 1ms. There is also an upper bound on the period length of 1s.
96 Additional restrictions exist when bandwidth l 96 Additional restrictions exist when bandwidth limits are used in a hierarchical
97 fashion, these are explained in more detail be 97 fashion, these are explained in more detail below.
98 98
99 Writing any negative value to cpu.cfs_quota_us 99 Writing any negative value to cpu.cfs_quota_us will remove the bandwidth limit
100 and return the group to an unconstrained state 100 and return the group to an unconstrained state once more.
101 101
102 A value of 0 for cpu.cfs_burst_us indicates th 102 A value of 0 for cpu.cfs_burst_us indicates that the group can not accumulate
103 any unused bandwidth. It makes the traditional 103 any unused bandwidth. It makes the traditional bandwidth control behavior for
104 CFS unchanged. Writing any (valid) positive va 104 CFS unchanged. Writing any (valid) positive value(s) no larger than
105 cpu.cfs_quota_us into cpu.cfs_burst_us will en 105 cpu.cfs_quota_us into cpu.cfs_burst_us will enact the cap on unused bandwidth
106 accumulation. 106 accumulation.
107 107
108 Any updates to a group's bandwidth specificati 108 Any updates to a group's bandwidth specification will result in it becoming
109 unthrottled if it is in a constrained state. 109 unthrottled if it is in a constrained state.
110 110
111 System wide settings 111 System wide settings
112 -------------------- 112 --------------------
113 For efficiency run-time is transferred between 113 For efficiency run-time is transferred between the global pool and CPU local
114 "silos" in a batch fashion. This greatly reduc 114 "silos" in a batch fashion. This greatly reduces global accounting pressure
115 on large systems. The amount transferred each 115 on large systems. The amount transferred each time such an update is required
116 is described as the "slice". 116 is described as the "slice".
117 117
118 This is tunable via procfs:: 118 This is tunable via procfs::
119 119
120 /proc/sys/kernel/sched_cfs_bandwidth_s 120 /proc/sys/kernel/sched_cfs_bandwidth_slice_us (default=5ms)
121 121
122 Larger slice values will reduce transfer overh 122 Larger slice values will reduce transfer overheads, while smaller values allow
123 for more fine-grained consumption. 123 for more fine-grained consumption.
124 124
125 Statistics 125 Statistics
126 ---------- 126 ----------
127 A group's bandwidth statistics are exported vi 127 A group's bandwidth statistics are exported via 5 fields in cpu.stat.
128 128
129 cpu.stat: 129 cpu.stat:
130 130
131 - nr_periods: Number of enforcement intervals 131 - nr_periods: Number of enforcement intervals that have elapsed.
132 - nr_throttled: Number of times the group has 132 - nr_throttled: Number of times the group has been throttled/limited.
133 - throttled_time: The total time duration (in 133 - throttled_time: The total time duration (in nanoseconds) for which entities
134 of the group have been throttled. 134 of the group have been throttled.
135 - nr_bursts: Number of periods burst occurs. 135 - nr_bursts: Number of periods burst occurs.
136 - burst_time: Cumulative wall-time (in nanosec 136 - burst_time: Cumulative wall-time (in nanoseconds) that any CPUs has used
137 above quota in respective periods. 137 above quota in respective periods.
138 138
139 This interface is read-only. 139 This interface is read-only.
140 140
141 Hierarchical considerations 141 Hierarchical considerations
142 --------------------------- 142 ---------------------------
143 The interface enforces that an individual enti 143 The interface enforces that an individual entity's bandwidth is always
144 attainable, that is: max(c_i) <= C. However, o 144 attainable, that is: max(c_i) <= C. However, over-subscription in the
145 aggregate case is explicitly allowed to enable 145 aggregate case is explicitly allowed to enable work-conserving semantics
146 within a hierarchy: 146 within a hierarchy:
147 147
148 e.g. \Sum (c_i) may exceed C 148 e.g. \Sum (c_i) may exceed C
149 149
150 [ Where C is the parent's bandwidth, and c_i i 150 [ Where C is the parent's bandwidth, and c_i its children ]
151 151
152 152
153 There are two ways in which a group may become 153 There are two ways in which a group may become throttled:
154 154
155 a. it fully consumes its own quota wit 155 a. it fully consumes its own quota within a period
156 b. a parent's quota is fully consumed 156 b. a parent's quota is fully consumed within its period
157 157
158 In case b) above, even though the child may ha 158 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 159 be allowed to until the parent's runtime is refreshed.
160 160
161 CFS Bandwidth Quota Caveats 161 CFS Bandwidth Quota Caveats
162 --------------------------- 162 ---------------------------
163 Once a slice is assigned to a cpu it does not 163 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 164 the slice may be returned to the global pool if all threads on that cpu become
165 unrunnable. This is configured at compile time 165 unrunnable. This is configured at compile time by the min_cfs_rq_runtime
166 variable. This is a performance tweak that hel 166 variable. This is a performance tweak that helps prevent added contention on
167 the global lock. 167 the global lock.
168 168
169 The fact that cpu-local slices do not expire r 169 The fact that cpu-local slices do not expire results in some interesting corner
170 cases that should be understood. 170 cases that should be understood.
171 171
172 For cgroup cpu constrained applications that a 172 For cgroup cpu constrained applications that are cpu limited this is a
173 relatively moot point because they will natura 173 relatively moot point because they will naturally consume the entirety of their
174 quota as well as the entirety of each cpu-loca 174 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 175 result it is expected that nr_periods roughly equal nr_throttled, and that
176 cpuacct.usage will increase roughly equal to c 176 cpuacct.usage will increase roughly equal to cfs_quota_us in each period.
177 177
178 For highly-threaded, non-cpu bound application 178 For highly-threaded, non-cpu bound applications this non-expiration nuance
179 allows applications to briefly burst past thei 179 allows applications to briefly burst past their quota limits by the amount of
180 unused slice on each cpu that the task group i 180 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 181 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 182 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 183 in previous periods. This burst amount will not be transferred between cores.
184 As a result, this mechanism still strictly lim 184 As a result, this mechanism still strictly limits the task group to quota
185 average usage, albeit over a longer time windo 185 average usage, albeit over a longer time window than a single period. This
186 also limits the burst ability to no more than 186 also limits the burst ability to no more than 1ms per cpu. This provides
187 better more predictable user experience for hi 187 better more predictable user experience for highly threaded applications with
188 small quota limits on high core count machines 188 small quota limits on high core count machines. It also eliminates the
189 propensity to throttle these applications whil 189 propensity to throttle these applications while simultaneously using less than
190 quota amounts of cpu. Another way to say this, 190 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 191 portion of a slice to remain valid across periods we have decreased the
192 possibility of wastefully expiring quota on cp 192 possibility of wastefully expiring quota on cpu-local silos that don't need a
193 full slice's amount of cpu time. 193 full slice's amount of cpu time.
194 194
195 The interaction between cpu-bound and non-cpu- 195 The interaction between cpu-bound and non-cpu-bound-interactive applications
196 should also be considered, especially when sin 196 should also be considered, especially when single core usage hits 100%. If you
197 gave each of these applications half of a cpu- 197 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 198 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 199 will use up to 1ms additional quota in some periods, thereby preventing the
200 cpu-bound application from fully using its quo 200 cpu-bound application from fully using its quota by that same amount. In these
201 instances it will be up to the CFS algorithm ( 201 instances it will be up to the CFS algorithm (see sched-design-CFS.rst) to
202 decide which application is chosen to run, as 202 decide which application is chosen to run, as they will both be runnable and
203 have remaining quota. This runtime discrepancy 203 have remaining quota. This runtime discrepancy will be made up in the following
204 periods when the interactive application idles 204 periods when the interactive application idles.
205 205
206 Examples 206 Examples
207 -------- 207 --------
208 1. Limit a group to 1 CPU worth of runtime:: 208 1. Limit a group to 1 CPU worth of runtime::
209 209
210 If period is 250ms and quota is also 2 210 If period is 250ms and quota is also 250ms, the group will get
211 1 CPU worth of runtime every 250ms. 211 1 CPU worth of runtime every 250ms.
212 212
213 # echo 250000 > cpu.cfs_quota_us /* qu 213 # echo 250000 > cpu.cfs_quota_us /* quota = 250ms */
214 # echo 250000 > cpu.cfs_period_us /* p 214 # echo 250000 > cpu.cfs_period_us /* period = 250ms */
215 215
216 2. Limit a group to 2 CPUs worth of runtime on 216 2. Limit a group to 2 CPUs worth of runtime on a multi-CPU machine
217 217
218 With 500ms period and 1000ms quota, the gro 218 With 500ms period and 1000ms quota, the group can get 2 CPUs worth of
219 runtime every 500ms:: 219 runtime every 500ms::
220 220
221 # echo 1000000 > cpu.cfs_quota_us /* q 221 # echo 1000000 > cpu.cfs_quota_us /* quota = 1000ms */
222 # echo 500000 > cpu.cfs_period_us /* p 222 # echo 500000 > cpu.cfs_period_us /* period = 500ms */
223 223
224 The larger period here allows for incr 224 The larger period here allows for increased burst capacity.
225 225
226 3. Limit a group to 20% of 1 CPU. 226 3. Limit a group to 20% of 1 CPU.
227 227
228 With 50ms period, 10ms quota will be equiva 228 With 50ms period, 10ms quota will be equivalent to 20% of 1 CPU::
229 229
230 # echo 10000 > cpu.cfs_quota_us /* quo 230 # echo 10000 > cpu.cfs_quota_us /* quota = 10ms */
231 # echo 50000 > cpu.cfs_period_us /* pe 231 # echo 50000 > cpu.cfs_period_us /* period = 50ms */
232 232
233 By using a small period here we are ensurin 233 By using a small period here we are ensuring a consistent latency
234 response at the expense of burst capacity. 234 response at the expense of burst capacity.
235 235
236 4. Limit a group to 40% of 1 CPU, and allow ac 236 4. Limit a group to 40% of 1 CPU, and allow accumulate up to 20% of 1 CPU
237 additionally, in case accumulation has been 237 additionally, in case accumulation has been done.
238 238
239 With 50ms period, 20ms quota will be equiva 239 With 50ms period, 20ms quota will be equivalent to 40% of 1 CPU.
240 And 10ms burst will be equivalent to 20% of 240 And 10ms burst will be equivalent to 20% of 1 CPU::
241 241
242 # echo 20000 > cpu.cfs_quota_us /* quo 242 # echo 20000 > cpu.cfs_quota_us /* quota = 20ms */
243 # echo 50000 > cpu.cfs_period_us /* pe 243 # echo 50000 > cpu.cfs_period_us /* period = 50ms */
244 # echo 10000 > cpu.cfs_burst_us /* bur 244 # echo 10000 > cpu.cfs_burst_us /* burst = 10ms */
245 245
246 Larger buffer setting (no larger than quota 246 Larger buffer setting (no larger than quota) allows greater burst capacity.
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.