About Kernel Documentation Linux Kernel Contact Linux Resources Linux Blog

Documentation / scheduler / sched-deadline.txt




Custom Search

Based on kernel version 4.13.3. Page generated on 2017-09-23 13:56 EST.

1				  Deadline Task Scheduling
2				  ------------------------
3	
4	CONTENTS
5	========
6	
7	 0. WARNING
8	 1. Overview
9	 2. Scheduling algorithm
10	   2.1 Main algorithm
11	   2.2 Bandwidth reclaiming
12	 3. Scheduling Real-Time Tasks
13	   3.1 Definitions
14	   3.2 Schedulability Analysis for Uniprocessor Systems
15	   3.3 Schedulability Analysis for Multiprocessor Systems
16	   3.4 Relationship with SCHED_DEADLINE Parameters
17	 4. Bandwidth management
18	   4.1 System-wide settings
19	   4.2 Task interface
20	   4.3 Default behavior
21	   4.4 Behavior of sched_yield()
22	 5. Tasks CPU affinity
23	   5.1 SCHED_DEADLINE and cpusets HOWTO
24	 6. Future plans
25	 A. Test suite
26	 B. Minimal main()
27	
28	
29	0. WARNING
30	==========
31	
32	 Fiddling with these settings can result in an unpredictable or even unstable
33	 system behavior. As for -rt (group) scheduling, it is assumed that root users
34	 know what they're doing.
35	
36	
37	1. Overview
38	===========
39	
40	 The SCHED_DEADLINE policy contained inside the sched_dl scheduling class is
41	 basically an implementation of the Earliest Deadline First (EDF) scheduling
42	 algorithm, augmented with a mechanism (called Constant Bandwidth Server, CBS)
43	 that makes it possible to isolate the behavior of tasks between each other.
44	
45	
46	2. Scheduling algorithm
47	==================
48	
49	2.1 Main algorithm
50	------------------
51	
52	 SCHED_DEADLINE uses three parameters, named "runtime", "period", and
53	 "deadline", to schedule tasks. A SCHED_DEADLINE task should receive
54	 "runtime" microseconds of execution time every "period" microseconds, and
55	 these "runtime" microseconds are available within "deadline" microseconds
56	 from the beginning of the period.  In order to implement this behavior,
57	 every time the task wakes up, the scheduler computes a "scheduling deadline"
58	 consistent with the guarantee (using the CBS[2,3] algorithm). Tasks are then
59	 scheduled using EDF[1] on these scheduling deadlines (the task with the
60	 earliest scheduling deadline is selected for execution). Notice that the
61	 task actually receives "runtime" time units within "deadline" if a proper
62	 "admission control" strategy (see Section "4. Bandwidth management") is used
63	 (clearly, if the system is overloaded this guarantee cannot be respected).
64	
65	 Summing up, the CBS[2,3] algorithm assigns scheduling deadlines to tasks so
66	 that each task runs for at most its runtime every period, avoiding any
67	 interference between different tasks (bandwidth isolation), while the EDF[1]
68	 algorithm selects the task with the earliest scheduling deadline as the one
69	 to be executed next. Thanks to this feature, tasks that do not strictly comply
70	 with the "traditional" real-time task model (see Section 3) can effectively
71	 use the new policy.
72	
73	 In more details, the CBS algorithm assigns scheduling deadlines to
74	 tasks in the following way:
75	
76	  - Each SCHED_DEADLINE task is characterized by the "runtime",
77	    "deadline", and "period" parameters;
78	
79	  - The state of the task is described by a "scheduling deadline", and
80	    a "remaining runtime". These two parameters are initially set to 0;
81	
82	  - When a SCHED_DEADLINE task wakes up (becomes ready for execution),
83	    the scheduler checks if
84	
85	                 remaining runtime                  runtime
86	        ----------------------------------    >    ---------
87	        scheduling deadline - current time           period
88	
89	    then, if the scheduling deadline is smaller than the current time, or
90	    this condition is verified, the scheduling deadline and the
91	    remaining runtime are re-initialized as
92	
93	         scheduling deadline = current time + deadline
94	         remaining runtime = runtime
95	
96	    otherwise, the scheduling deadline and the remaining runtime are
97	    left unchanged;
98	
99	  - When a SCHED_DEADLINE task executes for an amount of time t, its
100	    remaining runtime is decreased as
101	
102	         remaining runtime = remaining runtime - t
103	
104	    (technically, the runtime is decreased at every tick, or when the
105	    task is descheduled / preempted);
106	
107	  - When the remaining runtime becomes less or equal than 0, the task is
108	    said to be "throttled" (also known as "depleted" in real-time literature)
109	    and cannot be scheduled until its scheduling deadline. The "replenishment
110	    time" for this task (see next item) is set to be equal to the current
111	    value of the scheduling deadline;
112	
113	  - When the current time is equal to the replenishment time of a
114	    throttled task, the scheduling deadline and the remaining runtime are
115	    updated as
116	
117	         scheduling deadline = scheduling deadline + period
118	         remaining runtime = remaining runtime + runtime
119	
120	
121	2.2 Bandwidth reclaiming
122	------------------------
123	
124	 Bandwidth reclaiming for deadline tasks is based on the GRUB (Greedy
125	 Reclamation of Unused Bandwidth) algorithm [15, 16, 17] and it is enabled
126	 when flag SCHED_FLAG_RECLAIM is set.
127	
128	 The following diagram illustrates the state names for tasks handled by GRUB:
129	
130	                             ------------
131	                 (d)        |   Active   |
132	              ------------->|            |
133	              |             | Contending |
134	              |              ------------
135	              |                A      |
136	          ----------           |      |
137	         |          |          |      |
138	         | Inactive |          |(b)   | (a)
139	         |          |          |      |
140	          ----------           |      |
141	              A                |      V
142	              |              ------------
143	              |             |   Active   |
144	              --------------|     Non    |
145	                 (c)        | Contending |
146	                             ------------
147	
148	 A task can be in one of the following states:
149	
150	  - ActiveContending: if it is ready for execution (or executing);
151	
152	  - ActiveNonContending: if it just blocked and has not yet surpassed the 0-lag
153	    time;
154	
155	  - Inactive: if it is blocked and has surpassed the 0-lag time.
156	
157	 State transitions:
158	
159	  (a) When a task blocks, it does not become immediately inactive since its
160	      bandwidth cannot be immediately reclaimed without breaking the
161	      real-time guarantees. It therefore enters a transitional state called
162	      ActiveNonContending. The scheduler arms the "inactive timer" to fire at
163	      the 0-lag time, when the task's bandwidth can be reclaimed without
164	      breaking the real-time guarantees.
165	
166	      The 0-lag time for a task entering the ActiveNonContending state is
167	      computed as
168	
169	                        (runtime * dl_period)
170	             deadline - ---------------------
171	                             dl_runtime
172	
173	      where runtime is the remaining runtime, while dl_runtime and dl_period
174	      are the reservation parameters.
175	
176	  (b) If the task wakes up before the inactive timer fires, the task re-enters
177	      the ActiveContending state and the "inactive timer" is canceled.
178	      In addition, if the task wakes up on a different runqueue, then
179	      the task's utilization must be removed from the previous runqueue's active
180	      utilization and must be added to the new runqueue's active utilization.
181	      In order to avoid races between a task waking up on a runqueue while the
182	       "inactive timer" is running on a different CPU, the "dl_non_contending"
183	      flag is used to indicate that a task is not on a runqueue but is active
184	      (so, the flag is set when the task blocks and is cleared when the
185	      "inactive timer" fires or when the task  wakes up).
186	
187	  (c) When the "inactive timer" fires, the task enters the Inactive state and
188	      its utilization is removed from the runqueue's active utilization.
189	
190	  (d) When an inactive task wakes up, it enters the ActiveContending state and
191	      its utilization is added to the active utilization of the runqueue where
192	      it has been enqueued.
193	
194	 For each runqueue, the algorithm GRUB keeps track of two different bandwidths:
195	
196	  - Active bandwidth (running_bw): this is the sum of the bandwidths of all
197	    tasks in active state (i.e., ActiveContending or ActiveNonContending);
198	
199	  - Total bandwidth (this_bw): this is the sum of all tasks "belonging" to the
200	    runqueue, including the tasks in Inactive state.
201	
202	
203	 The algorithm reclaims the bandwidth of the tasks in Inactive state.
204	 It does so by decrementing the runtime of the executing task Ti at a pace equal
205	 to
206	
207	           dq = -max{ Ui, (1 - Uinact) } dt
208	
209	 where Uinact is the inactive utilization, computed as (this_bq - running_bw),
210	 and Ui is the bandwidth of task Ti.
211	
212	
213	 Let's now see a trivial example of two deadline tasks with runtime equal
214	 to 4 and period equal to 8 (i.e., bandwidth equal to 0.5):
215	
216	     A            Task T1
217	     |
218	     |                               |
219	     |                               |
220	     |--------                       |----
221	     |       |                       V
222	     |---|---|---|---|---|---|---|---|--------->t
223	     0   1   2   3   4   5   6   7   8
224	
225	
226	     A            Task T2
227	     |
228	     |                               |
229	     |                               |
230	     |       ------------------------|
231	     |       |                       V
232	     |---|---|---|---|---|---|---|---|--------->t
233	     0   1   2   3   4   5   6   7   8
234	
235	
236	     A            running_bw
237	     |
238	   1 -----------------               ------
239	     |               |               |
240	  0.5-               -----------------
241	     |                               |
242	     |---|---|---|---|---|---|---|---|--------->t
243	     0   1   2   3   4   5   6   7   8
244	
245	
246	  - Time t = 0:
247	
248	    Both tasks are ready for execution and therefore in ActiveContending state.
249	    Suppose Task T1 is the first task to start execution.
250	    Since there are no inactive tasks, its runtime is decreased as dq = -1 dt.
251	
252	  - Time t = 2:
253	
254	    Suppose that task T1 blocks
255	    Task T1 therefore enters the ActiveNonContending state. Since its remaining
256	    runtime is equal to 2, its 0-lag time is equal to t = 4.
257	    Task T2 start execution, with runtime still decreased as dq = -1 dt since
258	    there are no inactive tasks.
259	
260	  - Time t = 4:
261	
262	    This is the 0-lag time for Task T1. Since it didn't woken up in the
263	    meantime, it enters the Inactive state. Its bandwidth is removed from
264	    running_bw.
265	    Task T2 continues its execution. However, its runtime is now decreased as
266	    dq = - 0.5 dt because Uinact = 0.5.
267	    Task T2 therefore reclaims the bandwidth unused by Task T1.
268	
269	  - Time t = 8:
270	
271	    Task T1 wakes up. It enters the ActiveContending state again, and the
272	    running_bw is incremented.
273	
274	
275	3. Scheduling Real-Time Tasks
276	=============================
277	
278	 * BIG FAT WARNING ******************************************************
279	 *
280	 * This section contains a (not-thorough) summary on classical deadline
281	 * scheduling theory, and how it applies to SCHED_DEADLINE.
282	 * The reader can "safely" skip to Section 4 if only interested in seeing
283	 * how the scheduling policy can be used. Anyway, we strongly recommend
284	 * to come back here and continue reading (once the urge for testing is
285	 * satisfied :P) to be sure of fully understanding all technical details.
286	 ************************************************************************
287	
288	 There are no limitations on what kind of task can exploit this new
289	 scheduling discipline, even if it must be said that it is particularly
290	 suited for periodic or sporadic real-time tasks that need guarantees on their
291	 timing behavior, e.g., multimedia, streaming, control applications, etc.
292	
293	3.1 Definitions
294	------------------------
295	
296	 A typical real-time task is composed of a repetition of computation phases
297	 (task instances, or jobs) which are activated on a periodic or sporadic
298	 fashion.
299	 Each job J_j (where J_j is the j^th job of the task) is characterized by an
300	 arrival time r_j (the time when the job starts), an amount of computation
301	 time c_j needed to finish the job, and a job absolute deadline d_j, which
302	 is the time within which the job should be finished. The maximum execution
303	 time max{c_j} is called "Worst Case Execution Time" (WCET) for the task.
304	 A real-time task can be periodic with period P if r_{j+1} = r_j + P, or
305	 sporadic with minimum inter-arrival time P is r_{j+1} >= r_j + P. Finally,
306	 d_j = r_j + D, where D is the task's relative deadline.
307	 Summing up, a real-time task can be described as
308		Task = (WCET, D, P)
309	
310	 The utilization of a real-time task is defined as the ratio between its
311	 WCET and its period (or minimum inter-arrival time), and represents
312	 the fraction of CPU time needed to execute the task.
313	
314	 If the total utilization U=sum(WCET_i/P_i) is larger than M (with M equal
315	 to the number of CPUs), then the scheduler is unable to respect all the
316	 deadlines.
317	 Note that total utilization is defined as the sum of the utilizations
318	 WCET_i/P_i over all the real-time tasks in the system. When considering
319	 multiple real-time tasks, the parameters of the i-th task are indicated
320	 with the "_i" suffix.
321	 Moreover, if the total utilization is larger than M, then we risk starving
322	 non- real-time tasks by real-time tasks.
323	 If, instead, the total utilization is smaller than M, then non real-time
324	 tasks will not be starved and the system might be able to respect all the
325	 deadlines.
326	 As a matter of fact, in this case it is possible to provide an upper bound
327	 for tardiness (defined as the maximum between 0 and the difference
328	 between the finishing time of a job and its absolute deadline).
329	 More precisely, it can be proven that using a global EDF scheduler the
330	 maximum tardiness of each task is smaller or equal than
331		((M − 1) · WCET_max − WCET_min)/(M − (M − 2) · U_max) + WCET_max
332	 where WCET_max = max{WCET_i} is the maximum WCET, WCET_min=min{WCET_i}
333	 is the minimum WCET, and U_max = max{WCET_i/P_i} is the maximum
334	 utilization[12].
335	
336	3.2 Schedulability Analysis for Uniprocessor Systems
337	------------------------
338	
339	 If M=1 (uniprocessor system), or in case of partitioned scheduling (each
340	 real-time task is statically assigned to one and only one CPU), it is
341	 possible to formally check if all the deadlines are respected.
342	 If D_i = P_i for all tasks, then EDF is able to respect all the deadlines
343	 of all the tasks executing on a CPU if and only if the total utilization
344	 of the tasks running on such a CPU is smaller or equal than 1.
345	 If D_i != P_i for some task, then it is possible to define the density of
346	 a task as WCET_i/min{D_i,P_i}, and EDF is able to respect all the deadlines
347	 of all the tasks running on a CPU if the sum of the densities of the tasks
348	 running on such a CPU is smaller or equal than 1:
349		sum(WCET_i / min{D_i, P_i}) <= 1
350	 It is important to notice that this condition is only sufficient, and not
351	 necessary: there are task sets that are schedulable, but do not respect the
352	 condition. For example, consider the task set {Task_1,Task_2} composed by
353	 Task_1=(50ms,50ms,100ms) and Task_2=(10ms,100ms,100ms).
354	 EDF is clearly able to schedule the two tasks without missing any deadline
355	 (Task_1 is scheduled as soon as it is released, and finishes just in time
356	 to respect its deadline; Task_2 is scheduled immediately after Task_1, hence
357	 its response time cannot be larger than 50ms + 10ms = 60ms) even if
358		50 / min{50,100} + 10 / min{100, 100} = 50 / 50 + 10 / 100 = 1.1
359	 Of course it is possible to test the exact schedulability of tasks with
360	 D_i != P_i (checking a condition that is both sufficient and necessary),
361	 but this cannot be done by comparing the total utilization or density with
362	 a constant. Instead, the so called "processor demand" approach can be used,
363	 computing the total amount of CPU time h(t) needed by all the tasks to
364	 respect all of their deadlines in a time interval of size t, and comparing
365	 such a time with the interval size t. If h(t) is smaller than t (that is,
366	 the amount of time needed by the tasks in a time interval of size t is
367	 smaller than the size of the interval) for all the possible values of t, then
368	 EDF is able to schedule the tasks respecting all of their deadlines. Since
369	 performing this check for all possible values of t is impossible, it has been
370	 proven[4,5,6] that it is sufficient to perform the test for values of t
371	 between 0 and a maximum value L. The cited papers contain all of the
372	 mathematical details and explain how to compute h(t) and L.
373	 In any case, this kind of analysis is too complex as well as too
374	 time-consuming to be performed on-line. Hence, as explained in Section
375	 4 Linux uses an admission test based on the tasks' utilizations.
376	
377	3.3 Schedulability Analysis for Multiprocessor Systems
378	------------------------
379	
380	 On multiprocessor systems with global EDF scheduling (non partitioned
381	 systems), a sufficient test for schedulability can not be based on the
382	 utilizations or densities: it can be shown that even if D_i = P_i task
383	 sets with utilizations slightly larger than 1 can miss deadlines regardless
384	 of the number of CPUs.
385	
386	 Consider a set {Task_1,...Task_{M+1}} of M+1 tasks on a system with M
387	 CPUs, with the first task Task_1=(P,P,P) having period, relative deadline
388	 and WCET equal to P. The remaining M tasks Task_i=(e,P-1,P-1) have an
389	 arbitrarily small worst case execution time (indicated as "e" here) and a
390	 period smaller than the one of the first task. Hence, if all the tasks
391	 activate at the same time t, global EDF schedules these M tasks first
392	 (because their absolute deadlines are equal to t + P - 1, hence they are
393	 smaller than the absolute deadline of Task_1, which is t + P). As a
394	 result, Task_1 can be scheduled only at time t + e, and will finish at
395	 time t + e + P, after its absolute deadline. The total utilization of the
396	 task set is U = M · e / (P - 1) + P / P = M · e / (P - 1) + 1, and for small
397	 values of e this can become very close to 1. This is known as "Dhall's
398	 effect"[7]. Note: the example in the original paper by Dhall has been
399	 slightly simplified here (for example, Dhall more correctly computed
400	 lim_{e->0}U).
401	
402	 More complex schedulability tests for global EDF have been developed in
403	 real-time literature[8,9], but they are not based on a simple comparison
404	 between total utilization (or density) and a fixed constant. If all tasks
405	 have D_i = P_i, a sufficient schedulability condition can be expressed in
406	 a simple way:
407		sum(WCET_i / P_i) <= M - (M - 1) · U_max
408	 where U_max = max{WCET_i / P_i}[10]. Notice that for U_max = 1,
409	 M - (M - 1) · U_max becomes M - M + 1 = 1 and this schedulability condition
410	 just confirms the Dhall's effect. A more complete survey of the literature
411	 about schedulability tests for multi-processor real-time scheduling can be
412	 found in [11].
413	
414	 As seen, enforcing that the total utilization is smaller than M does not
415	 guarantee that global EDF schedules the tasks without missing any deadline
416	 (in other words, global EDF is not an optimal scheduling algorithm). However,
417	 a total utilization smaller than M is enough to guarantee that non real-time
418	 tasks are not starved and that the tardiness of real-time tasks has an upper
419	 bound[12] (as previously noted). Different bounds on the maximum tardiness
420	 experienced by real-time tasks have been developed in various papers[13,14],
421	 but the theoretical result that is important for SCHED_DEADLINE is that if
422	 the total utilization is smaller or equal than M then the response times of
423	 the tasks are limited.
424	
425	3.4 Relationship with SCHED_DEADLINE Parameters
426	------------------------
427	
428	 Finally, it is important to understand the relationship between the
429	 SCHED_DEADLINE scheduling parameters described in Section 2 (runtime,
430	 deadline and period) and the real-time task parameters (WCET, D, P)
431	 described in this section. Note that the tasks' temporal constraints are
432	 represented by its absolute deadlines d_j = r_j + D described above, while
433	 SCHED_DEADLINE schedules the tasks according to scheduling deadlines (see
434	 Section 2).
435	 If an admission test is used to guarantee that the scheduling deadlines
436	 are respected, then SCHED_DEADLINE can be used to schedule real-time tasks
437	 guaranteeing that all the jobs' deadlines of a task are respected.
438	 In order to do this, a task must be scheduled by setting:
439	
440	  - runtime >= WCET
441	  - deadline = D
442	  - period <= P
443	
444	 IOW, if runtime >= WCET and if period is <= P, then the scheduling deadlines
445	 and the absolute deadlines (d_j) coincide, so a proper admission control
446	 allows to respect the jobs' absolute deadlines for this task (this is what is
447	 called "hard schedulability property" and is an extension of Lemma 1 of [2]).
448	 Notice that if runtime > deadline the admission control will surely reject
449	 this task, as it is not possible to respect its temporal constraints.
450	
451	 References:
452	  1 - C. L. Liu and J. W. Layland. Scheduling algorithms for multiprogram-
453	      ming in a hard-real-time environment. Journal of the Association for
454	      Computing Machinery, 20(1), 1973.
455	  2 - L. Abeni , G. Buttazzo. Integrating Multimedia Applications in Hard
456	      Real-Time Systems. Proceedings of the 19th IEEE Real-time Systems
457	      Symposium, 1998. http://retis.sssup.it/~giorgio/paps/1998/rtss98-cbs.pdf
458	  3 - L. Abeni. Server Mechanisms for Multimedia Applications. ReTiS Lab
459	      Technical Report. http://disi.unitn.it/~abeni/tr-98-01.pdf
460	  4 - J. Y. Leung and M.L. Merril. A Note on Preemptive Scheduling of
461	      Periodic, Real-Time Tasks. Information Processing Letters, vol. 11,
462	      no. 3, pp. 115-118, 1980.
463	  5 - S. K. Baruah, A. K. Mok and L. E. Rosier. Preemptively Scheduling
464	      Hard-Real-Time Sporadic Tasks on One Processor. Proceedings of the
465	      11th IEEE Real-time Systems Symposium, 1990.
466	  6 - S. K. Baruah, L. E. Rosier and R. R. Howell. Algorithms and Complexity
467	      Concerning the Preemptive Scheduling of Periodic Real-Time tasks on
468	      One Processor. Real-Time Systems Journal, vol. 4, no. 2, pp 301-324,
469	      1990.
470	  7 - S. J. Dhall and C. L. Liu. On a real-time scheduling problem. Operations
471	      research, vol. 26, no. 1, pp 127-140, 1978.
472	  8 - T. Baker. Multiprocessor EDF and Deadline Monotonic Schedulability
473	      Analysis. Proceedings of the 24th IEEE Real-Time Systems Symposium, 2003.
474	  9 - T. Baker. An Analysis of EDF Schedulability on a Multiprocessor.
475	      IEEE Transactions on Parallel and Distributed Systems, vol. 16, no. 8,
476	      pp 760-768, 2005.
477	  10 - J. Goossens, S. Funk and S. Baruah, Priority-Driven Scheduling of
478	       Periodic Task Systems on Multiprocessors. Real-Time Systems Journal,
479	       vol. 25, no. 2–3, pp. 187–205, 2003.
480	  11 - R. Davis and A. Burns. A Survey of Hard Real-Time Scheduling for
481	       Multiprocessor Systems. ACM Computing Surveys, vol. 43, no. 4, 2011.
482	       http://www-users.cs.york.ac.uk/~robdavis/papers/MPSurveyv5.0.pdf
483	  12 - U. C. Devi and J. H. Anderson. Tardiness Bounds under Global EDF
484	       Scheduling on a Multiprocessor. Real-Time Systems Journal, vol. 32,
485	       no. 2, pp 133-189, 2008.
486	  13 - P. Valente and G. Lipari. An Upper Bound to the Lateness of Soft
487	       Real-Time Tasks Scheduled by EDF on Multiprocessors. Proceedings of
488	       the 26th IEEE Real-Time Systems Symposium, 2005.
489	  14 - J. Erickson, U. Devi and S. Baruah. Improved tardiness bounds for
490	       Global EDF. Proceedings of the 22nd Euromicro Conference on
491	       Real-Time Systems, 2010.
492	  15 - G. Lipari, S. Baruah, Greedy reclamation of unused bandwidth in
493	       constant-bandwidth servers, 12th IEEE Euromicro Conference on Real-Time
494	       Systems, 2000.
495	  16 - L. Abeni, J. Lelli, C. Scordino, L. Palopoli, Greedy CPU reclaiming for
496	       SCHED DEADLINE. In Proceedings of the Real-Time Linux Workshop (RTLWS),
497	       Dusseldorf, Germany, 2014.
498	  17 - L. Abeni, G. Lipari, A. Parri, Y. Sun, Multicore CPU reclaiming: parallel
499	       or sequential?. In Proceedings of the 31st Annual ACM Symposium on Applied
500	       Computing, 2016.
501	
502	
503	4. Bandwidth management
504	=======================
505	
506	 As previously mentioned, in order for -deadline scheduling to be
507	 effective and useful (that is, to be able to provide "runtime" time units
508	 within "deadline"), it is important to have some method to keep the allocation
509	 of the available fractions of CPU time to the various tasks under control.
510	 This is usually called "admission control" and if it is not performed, then
511	 no guarantee can be given on the actual scheduling of the -deadline tasks.
512	
513	 As already stated in Section 3, a necessary condition to be respected to
514	 correctly schedule a set of real-time tasks is that the total utilization
515	 is smaller than M. When talking about -deadline tasks, this requires that
516	 the sum of the ratio between runtime and period for all tasks is smaller
517	 than M. Notice that the ratio runtime/period is equivalent to the utilization
518	 of a "traditional" real-time task, and is also often referred to as
519	 "bandwidth".
520	 The interface used to control the CPU bandwidth that can be allocated
521	 to -deadline tasks is similar to the one already used for -rt
522	 tasks with real-time group scheduling (a.k.a. RT-throttling - see
523	 Documentation/scheduler/sched-rt-group.txt), and is based on readable/
524	 writable control files located in procfs (for system wide settings).
525	 Notice that per-group settings (controlled through cgroupfs) are still not
526	 defined for -deadline tasks, because more discussion is needed in order to
527	 figure out how we want to manage SCHED_DEADLINE bandwidth at the task group
528	 level.
529	
530	 A main difference between deadline bandwidth management and RT-throttling
531	 is that -deadline tasks have bandwidth on their own (while -rt ones don't!),
532	 and thus we don't need a higher level throttling mechanism to enforce the
533	 desired bandwidth. In other words, this means that interface parameters are
534	 only used at admission control time (i.e., when the user calls
535	 sched_setattr()). Scheduling is then performed considering actual tasks'
536	 parameters, so that CPU bandwidth is allocated to SCHED_DEADLINE tasks
537	 respecting their needs in terms of granularity. Therefore, using this simple
538	 interface we can put a cap on total utilization of -deadline tasks (i.e.,
539	 \Sum (runtime_i / period_i) < global_dl_utilization_cap).
540	
541	4.1 System wide settings
542	------------------------
543	
544	 The system wide settings are configured under the /proc virtual file system.
545	
546	 For now the -rt knobs are used for -deadline admission control and the
547	 -deadline runtime is accounted against the -rt runtime. We realize that this
548	 isn't entirely desirable; however, it is better to have a small interface for
549	 now, and be able to change it easily later. The ideal situation (see 5.) is to
550	 run -rt tasks from a -deadline server; in which case the -rt bandwidth is a
551	 direct subset of dl_bw.
552	
553	 This means that, for a root_domain comprising M CPUs, -deadline tasks
554	 can be created while the sum of their bandwidths stays below:
555	
556	   M * (sched_rt_runtime_us / sched_rt_period_us)
557	
558	 It is also possible to disable this bandwidth management logic, and
559	 be thus free of oversubscribing the system up to any arbitrary level.
560	 This is done by writing -1 in /proc/sys/kernel/sched_rt_runtime_us.
561	
562	
563	4.2 Task interface
564	------------------
565	
566	 Specifying a periodic/sporadic task that executes for a given amount of
567	 runtime at each instance, and that is scheduled according to the urgency of
568	 its own timing constraints needs, in general, a way of declaring:
569	  - a (maximum/typical) instance execution time,
570	  - a minimum interval between consecutive instances,
571	  - a time constraint by which each instance must be completed.
572	
573	 Therefore:
574	  * a new struct sched_attr, containing all the necessary fields is
575	    provided;
576	  * the new scheduling related syscalls that manipulate it, i.e.,
577	    sched_setattr() and sched_getattr() are implemented.
578	
579	 For debugging purposes, the leftover runtime and absolute deadline of a
580	 SCHED_DEADLINE task can be retrieved through /proc/<pid>/sched (entries
581	 dl.runtime and dl.deadline, both values in ns). A programmatic way to
582	 retrieve these values from production code is under discussion.
583	
584	
585	4.3 Default behavior
586	---------------------
587	
588	 The default value for SCHED_DEADLINE bandwidth is to have rt_runtime equal to
589	 950000. With rt_period equal to 1000000, by default, it means that -deadline
590	 tasks can use at most 95%, multiplied by the number of CPUs that compose the
591	 root_domain, for each root_domain.
592	 This means that non -deadline tasks will receive at least 5% of the CPU time,
593	 and that -deadline tasks will receive their runtime with a guaranteed
594	 worst-case delay respect to the "deadline" parameter. If "deadline" = "period"
595	 and the cpuset mechanism is used to implement partitioned scheduling (see
596	 Section 5), then this simple setting of the bandwidth management is able to
597	 deterministically guarantee that -deadline tasks will receive their runtime
598	 in a period.
599	
600	 Finally, notice that in order not to jeopardize the admission control a
601	 -deadline task cannot fork.
602	
603	
604	4.4 Behavior of sched_yield()
605	-----------------------------
606	
607	 When a SCHED_DEADLINE task calls sched_yield(), it gives up its
608	 remaining runtime and is immediately throttled, until the next
609	 period, when its runtime will be replenished (a special flag
610	 dl_yielded is set and used to handle correctly throttling and runtime
611	 replenishment after a call to sched_yield()).
612	
613	 This behavior of sched_yield() allows the task to wake-up exactly at
614	 the beginning of the next period. Also, this may be useful in the
615	 future with bandwidth reclaiming mechanisms, where sched_yield() will
616	 make the leftoever runtime available for reclamation by other
617	 SCHED_DEADLINE tasks.
618	
619	
620	5. Tasks CPU affinity
621	=====================
622	
623	 -deadline tasks cannot have an affinity mask smaller that the entire
624	 root_domain they are created on. However, affinities can be specified
625	 through the cpuset facility (Documentation/cgroup-v1/cpusets.txt).
626	
627	5.1 SCHED_DEADLINE and cpusets HOWTO
628	------------------------------------
629	
630	 An example of a simple configuration (pin a -deadline task to CPU0)
631	 follows (rt-app is used to create a -deadline task).
632	
633	 mkdir /dev/cpuset
634	 mount -t cgroup -o cpuset cpuset /dev/cpuset
635	 cd /dev/cpuset
636	 mkdir cpu0
637	 echo 0 > cpu0/cpuset.cpus
638	 echo 0 > cpu0/cpuset.mems
639	 echo 1 > cpuset.cpu_exclusive
640	 echo 0 > cpuset.sched_load_balance
641	 echo 1 > cpu0/cpuset.cpu_exclusive
642	 echo 1 > cpu0/cpuset.mem_exclusive
643	 echo $$ > cpu0/tasks
644	 rt-app -t 100000:10000:d:0 -D5 (it is now actually superfluous to specify
645	 task affinity)
646	
647	6. Future plans
648	===============
649	
650	 Still missing:
651	
652	  - programmatic way to retrieve current runtime and absolute deadline
653	  - refinements to deadline inheritance, especially regarding the possibility
654	    of retaining bandwidth isolation among non-interacting tasks. This is
655	    being studied from both theoretical and practical points of view, and
656	    hopefully we should be able to produce some demonstrative code soon;
657	  - (c)group based bandwidth management, and maybe scheduling;
658	  - access control for non-root users (and related security concerns to
659	    address), which is the best way to allow unprivileged use of the mechanisms
660	    and how to prevent non-root users "cheat" the system?
661	
662	 As already discussed, we are planning also to merge this work with the EDF
663	 throttling patches [https://lkml.org/lkml/2010/2/23/239] but we still are in
664	 the preliminary phases of the merge and we really seek feedback that would
665	 help us decide on the direction it should take.
666	
667	Appendix A. Test suite
668	======================
669	
670	 The SCHED_DEADLINE policy can be easily tested using two applications that
671	 are part of a wider Linux Scheduler validation suite. The suite is
672	 available as a GitHub repository: https://github.com/scheduler-tools.
673	
674	 The first testing application is called rt-app and can be used to
675	 start multiple threads with specific parameters. rt-app supports
676	 SCHED_{OTHER,FIFO,RR,DEADLINE} scheduling policies and their related
677	 parameters (e.g., niceness, priority, runtime/deadline/period). rt-app
678	 is a valuable tool, as it can be used to synthetically recreate certain
679	 workloads (maybe mimicking real use-cases) and evaluate how the scheduler
680	 behaves under such workloads. In this way, results are easily reproducible.
681	 rt-app is available at: https://github.com/scheduler-tools/rt-app.
682	
683	 Thread parameters can be specified from the command line, with something like
684	 this:
685	
686	  # rt-app -t 100000:10000:d -t 150000:20000:f:10 -D5
687	
688	 The above creates 2 threads. The first one, scheduled by SCHED_DEADLINE,
689	 executes for 10ms every 100ms. The second one, scheduled at SCHED_FIFO
690	 priority 10, executes for 20ms every 150ms. The test will run for a total
691	 of 5 seconds.
692	
693	 More interestingly, configurations can be described with a json file that
694	 can be passed as input to rt-app with something like this:
695	
696	  # rt-app my_config.json
697	
698	 The parameters that can be specified with the second method are a superset
699	 of the command line options. Please refer to rt-app documentation for more
700	 details (<rt-app-sources>/doc/*.json).
701	
702	 The second testing application is a modification of schedtool, called
703	 schedtool-dl, which can be used to setup SCHED_DEADLINE parameters for a
704	 certain pid/application. schedtool-dl is available at:
705	 https://github.com/scheduler-tools/schedtool-dl.git.
706	
707	 The usage is straightforward:
708	
709	  # schedtool -E -t 10000000:100000000 -e ./my_cpuhog_app
710	
711	 With this, my_cpuhog_app is put to run inside a SCHED_DEADLINE reservation
712	 of 10ms every 100ms (note that parameters are expressed in microseconds).
713	 You can also use schedtool to create a reservation for an already running
714	 application, given that you know its pid:
715	
716	  # schedtool -E -t 10000000:100000000 my_app_pid
717	
718	Appendix B. Minimal main()
719	==========================
720	
721	 We provide in what follows a simple (ugly) self-contained code snippet
722	 showing how SCHED_DEADLINE reservations can be created by a real-time
723	 application developer.
724	
725	 #define _GNU_SOURCE
726	 #include <unistd.h>
727	 #include <stdio.h>
728	 #include <stdlib.h>
729	 #include <string.h>
730	 #include <time.h>
731	 #include <linux/unistd.h>
732	 #include <linux/kernel.h>
733	 #include <linux/types.h>
734	 #include <sys/syscall.h>
735	 #include <pthread.h>
736	
737	 #define gettid() syscall(__NR_gettid)
738	
739	 #define SCHED_DEADLINE	6
740	
741	 /* XXX use the proper syscall numbers */
742	 #ifdef __x86_64__
743	 #define __NR_sched_setattr		314
744	 #define __NR_sched_getattr		315
745	 #endif
746	
747	 #ifdef __i386__
748	 #define __NR_sched_setattr		351
749	 #define __NR_sched_getattr		352
750	 #endif
751	
752	 #ifdef __arm__
753	 #define __NR_sched_setattr		380
754	 #define __NR_sched_getattr		381
755	 #endif
756	
757	 static volatile int done;
758	
759	 struct sched_attr {
760		__u32 size;
761	
762		__u32 sched_policy;
763		__u64 sched_flags;
764	
765		/* SCHED_NORMAL, SCHED_BATCH */
766		__s32 sched_nice;
767	
768		/* SCHED_FIFO, SCHED_RR */
769		__u32 sched_priority;
770	
771		/* SCHED_DEADLINE (nsec) */
772		__u64 sched_runtime;
773		__u64 sched_deadline;
774		__u64 sched_period;
775	 };
776	
777	 int sched_setattr(pid_t pid,
778			  const struct sched_attr *attr,
779			  unsigned int flags)
780	 {
781		return syscall(__NR_sched_setattr, pid, attr, flags);
782	 }
783	
784	 int sched_getattr(pid_t pid,
785			  struct sched_attr *attr,
786			  unsigned int size,
787			  unsigned int flags)
788	 {
789		return syscall(__NR_sched_getattr, pid, attr, size, flags);
790	 }
791	
792	 void *run_deadline(void *data)
793	 {
794		struct sched_attr attr;
795		int x = 0;
796		int ret;
797		unsigned int flags = 0;
798	
799		printf("deadline thread started [%ld]\n", gettid());
800	
801		attr.size = sizeof(attr);
802		attr.sched_flags = 0;
803		attr.sched_nice = 0;
804		attr.sched_priority = 0;
805	
806		/* This creates a 10ms/30ms reservation */
807		attr.sched_policy = SCHED_DEADLINE;
808		attr.sched_runtime = 10 * 1000 * 1000;
809		attr.sched_period = attr.sched_deadline = 30 * 1000 * 1000;
810	
811		ret = sched_setattr(0, &attr, flags);
812		if (ret < 0) {
813			done = 0;
814			perror("sched_setattr");
815			exit(-1);
816		}
817	
818		while (!done) {
819			x++;
820		}
821	
822		printf("deadline thread dies [%ld]\n", gettid());
823		return NULL;
824	 }
825	
826	 int main (int argc, char **argv)
827	 {
828		pthread_t thread;
829	
830		printf("main thread [%ld]\n", gettid());
831	
832		pthread_create(&thread, NULL, run_deadline, NULL);
833	
834		sleep(10);
835	
836		done = 1;
837		pthread_join(thread, NULL);
838	
839		printf("main dies [%ld]\n", gettid());
840		return 0;
841	 }
Hide Line Numbers
About Kernel Documentation Linux Kernel Contact Linux Resources Linux Blog

Information is copyright its respective author. All material is available from the Linux Kernel Source distributed under a GPL License. This page is provided as a free service by mjmwired.net.