blob: 18adc92a6b3b9ecac115965443d622128f2467e4 [file] [log] [blame]
Dario Faggioli712e5e32014-01-27 12:20:15 +01001 Deadline Task Scheduling
2 ------------------------
3
4CONTENTS
5========
6
7 0. WARNING
8 1. Overview
9 2. Scheduling algorithm
10 3. Scheduling Real-Time Tasks
11 4. Bandwidth management
12 4.1 System-wide settings
13 4.2 Task interface
14 4.3 Default behavior
15 5. Tasks CPU affinity
16 5.1 SCHED_DEADLINE and cpusets HOWTO
17 6. Future plans
18
19
200. WARNING
21==========
22
23 Fiddling with these settings can result in an unpredictable or even unstable
24 system behavior. As for -rt (group) scheduling, it is assumed that root users
25 know what they're doing.
26
27
281. Overview
29===========
30
31 The SCHED_DEADLINE policy contained inside the sched_dl scheduling class is
32 basically an implementation of the Earliest Deadline First (EDF) scheduling
33 algorithm, augmented with a mechanism (called Constant Bandwidth Server, CBS)
34 that makes it possible to isolate the behavior of tasks between each other.
35
36
372. Scheduling algorithm
38==================
39
40 SCHED_DEADLINE uses three parameters, named "runtime", "period", and
41 "deadline" to schedule tasks. A SCHED_DEADLINE task is guaranteed to receive
42 "runtime" microseconds of execution time every "period" microseconds, and
43 these "runtime" microseconds are available within "deadline" microseconds
44 from the beginning of the period. In order to implement this behaviour,
45 every time the task wakes up, the scheduler computes a "scheduling deadline"
46 consistent with the guarantee (using the CBS[2,3] algorithm). Tasks are then
47 scheduled using EDF[1] on these scheduling deadlines (the task with the
48 smallest scheduling deadline is selected for execution). Notice that this
49 guaranteed is respected if a proper "admission control" strategy (see Section
50 "4. Bandwidth management") is used.
51
52 Summing up, the CBS[2,3] algorithms assigns scheduling deadlines to tasks so
53 that each task runs for at most its runtime every period, avoiding any
54 interference between different tasks (bandwidth isolation), while the EDF[1]
55 algorithm selects the task with the smallest scheduling deadline as the one
56 to be executed first. Thanks to this feature, also tasks that do not
57 strictly comply with the "traditional" real-time task model (see Section 3)
58 can effectively use the new policy.
59
60 In more details, the CBS algorithm assigns scheduling deadlines to
61 tasks in the following way:
62
63 - Each SCHED_DEADLINE task is characterised by the "runtime",
64 "deadline", and "period" parameters;
65
66 - The state of the task is described by a "scheduling deadline", and
67 a "current runtime". These two parameters are initially set to 0;
68
69 - When a SCHED_DEADLINE task wakes up (becomes ready for execution),
70 the scheduler checks if
71
72 current runtime runtime
73 ---------------------------------- > ----------------
74 scheduling deadline - current time period
75
76 then, if the scheduling deadline is smaller than the current time, or
77 this condition is verified, the scheduling deadline and the
78 current budget are re-initialised as
79
80 scheduling deadline = current time + deadline
81 current runtime = runtime
82
83 otherwise, the scheduling deadline and the current runtime are
84 left unchanged;
85
86 - When a SCHED_DEADLINE task executes for an amount of time t, its
87 current runtime is decreased as
88
89 current runtime = current runtime - t
90
91 (technically, the runtime is decreased at every tick, or when the
92 task is descheduled / preempted);
93
94 - When the current runtime becomes less or equal than 0, the task is
95 said to be "throttled" (also known as "depleted" in real-time literature)
96 and cannot be scheduled until its scheduling deadline. The "replenishment
97 time" for this task (see next item) is set to be equal to the current
98 value of the scheduling deadline;
99
100 - When the current time is equal to the replenishment time of a
101 throttled task, the scheduling deadline and the current runtime are
102 updated as
103
104 scheduling deadline = scheduling deadline + period
105 current runtime = current runtime + runtime
106
107
1083. Scheduling Real-Time Tasks
109=============================
110
111 * BIG FAT WARNING ******************************************************
112 *
113 * This section contains a (not-thorough) summary on classical deadline
114 * scheduling theory, and how it applies to SCHED_DEADLINE.
115 * The reader can "safely" skip to Section 4 if only interested in seeing
116 * how the scheduling policy can be used. Anyway, we strongly recommend
117 * to come back here and continue reading (once the urge for testing is
118 * satisfied :P) to be sure of fully understanding all technical details.
119 ************************************************************************
120
121 There are no limitations on what kind of task can exploit this new
122 scheduling discipline, even if it must be said that it is particularly
123 suited for periodic or sporadic real-time tasks that need guarantees on their
124 timing behavior, e.g., multimedia, streaming, control applications, etc.
125
126 A typical real-time task is composed of a repetition of computation phases
127 (task instances, or jobs) which are activated on a periodic or sporadic
128 fashion.
129 Each job J_j (where J_j is the j^th job of the task) is characterised by an
130 arrival time r_j (the time when the job starts), an amount of computation
131 time c_j needed to finish the job, and a job absolute deadline d_j, which
132 is the time within which the job should be finished. The maximum execution
133 time max_j{c_j} is called "Worst Case Execution Time" (WCET) for the task.
134 A real-time task can be periodic with period P if r_{j+1} = r_j + P, or
135 sporadic with minimum inter-arrival time P is r_{j+1} >= r_j + P. Finally,
136 d_j = r_j + D, where D is the task's relative deadline.
137
138 SCHED_DEADLINE can be used to schedule real-time tasks guaranteeing that
139 the jobs' deadlines of a task are respected. In order to do this, a task
140 must be scheduled by setting:
141
142 - runtime >= WCET
143 - deadline = D
144 - period <= P
145
146 IOW, if runtime >= WCET and if period is >= P, then the scheduling deadlines
147 and the absolute deadlines (d_j) coincide, so a proper admission control
148 allows to respect the jobs' absolute deadlines for this task (this is what is
149 called "hard schedulability property" and is an extension of Lemma 1 of [2]).
150
151 References:
152 1 - C. L. Liu and J. W. Layland. Scheduling algorithms for multiprogram-
153 ming in a hard-real-time environment. Journal of the Association for
154 Computing Machinery, 20(1), 1973.
155 2 - L. Abeni , G. Buttazzo. Integrating Multimedia Applications in Hard
156 Real-Time Systems. Proceedings of the 19th IEEE Real-time Systems
157 Symposium, 1998. http://retis.sssup.it/~giorgio/paps/1998/rtss98-cbs.pdf
158 3 - L. Abeni. Server Mechanisms for Multimedia Applications. ReTiS Lab
159 Technical Report. http://xoomer.virgilio.it/lucabe72/pubs/tr-98-01.ps
160
1614. Bandwidth management
162=======================
163
164 In order for the -deadline scheduling to be effective and useful, it is
165 important to have some method to keep the allocation of the available CPU
166 bandwidth to the tasks under control.
167 This is usually called "admission control" and if it is not performed at all,
168 no guarantee can be given on the actual scheduling of the -deadline tasks.
169
170 Since when RT-throttling has been introduced each task group has a bandwidth
171 associated, calculated as a certain amount of runtime over a period.
172 Moreover, to make it possible to manipulate such bandwidth, readable/writable
173 controls have been added to both procfs (for system wide settings) and cgroupfs
174 (for per-group settings).
175 Therefore, the same interface is being used for controlling the bandwidth
176 distrubution to -deadline tasks.
177
178 However, more discussion is needed in order to figure out how we want to manage
179 SCHED_DEADLINE bandwidth at the task group level. Therefore, SCHED_DEADLINE
180 uses (for now) a less sophisticated, but actually very sensible, mechanism to
181 ensure that a certain utilization cap is not overcome per each root_domain.
182
183 Another main difference between deadline bandwidth management and RT-throttling
184 is that -deadline tasks have bandwidth on their own (while -rt ones don't!),
185 and thus we don't need an higher level throttling mechanism to enforce the
186 desired bandwidth.
187
1884.1 System wide settings
189------------------------
190
191 The system wide settings are configured under the /proc virtual file system.
192
193 For now the -rt knobs are used for dl admission control and the -deadline
194 runtime is accounted against the -rt runtime. We realise that this isn't
195 entirely desirable; however, it is better to have a small interface for now,
196 and be able to change it easily later. The ideal situation (see 5.) is to run
197 -rt tasks from a -deadline server; in which case the -rt bandwidth is a direct
198 subset of dl_bw.
199
200 This means that, for a root_domain comprising M CPUs, -deadline tasks
201 can be created while the sum of their bandwidths stays below:
202
203 M * (sched_rt_runtime_us / sched_rt_period_us)
204
205 It is also possible to disable this bandwidth management logic, and
206 be thus free of oversubscribing the system up to any arbitrary level.
207 This is done by writing -1 in /proc/sys/kernel/sched_rt_runtime_us.
208
209
2104.2 Task interface
211------------------
212
213 Specifying a periodic/sporadic task that executes for a given amount of
214 runtime at each instance, and that is scheduled according to the urgency of
215 its own timing constraints needs, in general, a way of declaring:
216 - a (maximum/typical) instance execution time,
217 - a minimum interval between consecutive instances,
218 - a time constraint by which each instance must be completed.
219
220 Therefore:
221 * a new struct sched_attr, containing all the necessary fields is
222 provided;
223 * the new scheduling related syscalls that manipulate it, i.e.,
224 sched_setattr() and sched_getattr() are implemented.
225
226
2274.3 Default behavior
228---------------------
229
230 The default value for SCHED_DEADLINE bandwidth is to have rt_runtime equal to
231 950000. With rt_period equal to 1000000, by default, it means that -deadline
232 tasks can use at most 95%, multiplied by the number of CPUs that compose the
233 root_domain, for each root_domain.
234
235 A -deadline task cannot fork.
236
2375. Tasks CPU affinity
238=====================
239
240 -deadline tasks cannot have an affinity mask smaller that the entire
241 root_domain they are created on. However, affinities can be specified
242 through the cpuset facility (Documentation/cgroups/cpusets.txt).
243
2445.1 SCHED_DEADLINE and cpusets HOWTO
245------------------------------------
246
247 An example of a simple configuration (pin a -deadline task to CPU0)
248 follows (rt-app is used to create a -deadline task).
249
250 mkdir /dev/cpuset
251 mount -t cgroup -o cpuset cpuset /dev/cpuset
252 cd /dev/cpuset
253 mkdir cpu0
254 echo 0 > cpu0/cpuset.cpus
255 echo 0 > cpu0/cpuset.mems
256 echo 1 > cpuset.cpu_exclusive
257 echo 0 > cpuset.sched_load_balance
258 echo 1 > cpu0/cpuset.cpu_exclusive
259 echo 1 > cpu0/cpuset.mem_exclusive
260 echo $$ > cpu0/tasks
261 rt-app -t 100000:10000:d:0 -D5 (it is now actually superfluous to specify
262 task affinity)
263
2646. Future plans
265===============
266
267 Still missing:
268
269 - refinements to deadline inheritance, especially regarding the possibility
270 of retaining bandwidth isolation among non-interacting tasks. This is
271 being studied from both theoretical and practical points of view, and
272 hopefully we should be able to produce some demonstrative code soon;
273 - (c)group based bandwidth management, and maybe scheduling;
274 - access control for non-root users (and related security concerns to
275 address), which is the best way to allow unprivileged use of the mechanisms
276 and how to prevent non-root users "cheat" the system?
277
278 As already discussed, we are planning also to merge this work with the EDF
279 throttling patches [https://lkml.org/lkml/2010/2/23/239] but we still are in
280 the preliminary phases of the merge and we really seek feedback that would
281 help us decide on the direction it should take.