Documentation/scheduler/sched-design-CFS.txt

   1                       =============
   2                       CFS Scheduler
   3                       =============
   4
   5
   6 1.  OVERVIEW
   7
   8 CFS stands for "Completely Fair Scheduler," and is the new "desktop" process
   9 scheduler implemented by Ingo Molnar and merged in Linux 2.6.23.  It is the
  10 replacement for the previous vanilla scheduler's SCHED_OTHER interactivity
  11 code.
  12
  13 80% of CFS's design can be summed up in a single sentence: CFS basically models
  14 an "ideal, precise multi-tasking CPU" on real hardware.
  15
  16 "Ideal multi-tasking CPU" is a (non-existent  :-)) CPU that has 100% physical
  17 power and which can run each task at precise equal speed, in parallel, each at
  18 1/nr_running speed.  For example: if there are 2 tasks running, then it runs
  19 each at 50% physical power --- i.e., actually in parallel.
  20
  21 On real hardware, we can run only a single task at once, so we have to
  22 introduce the concept of "virtual runtime."  The virtual runtime of a task
  23 specifies when its next timeslice would start execution on the ideal
  24 multi-tasking CPU described above.  In practice, the virtual runtime of a task
  25 is its actual runtime normalized to the total number of running tasks.
  26
  27
  28
  29 2.  FEW IMPLEMENTATION DETAILS
  30
  31 In CFS the virtual runtime is expressed and tracked via the per-task
  32 p->se.vruntime (nanosec-unit) value.  This way, it's possible to accurately
  33 timestamp and measure the "expected CPU time" a task should have gotten.
  34
  35 [ small detail: on "ideal" hardware, at any time all tasks would have the same
  36   p->se.vruntime value --- i.e., tasks would execute simultaneously and no task
  37   would ever get "out of balance" from the "ideal" share of CPU time.  ]
  38
  39 CFS's task picking logic is based on this p->se.vruntime value and it is thus
  40 very simple: it always tries to run the task with the smallest p->se.vruntime
  41 value (i.e., the task which executed least so far).  CFS always tries to split
  42 up CPU time between runnable tasks as close to "ideal multitasking hardware" as
  43 possible.
  44
  45 Most of the rest of CFS's design just falls out of this really simple concept,
  46 with a few add-on embellishments like nice levels, multiprocessing and various
  47 algorithm variants to recognize sleepers.
  48
  49
  50
  51 3.  THE RBTREE
  52
  53 CFS's design is quite radical: it does not use the old data structures for the
  54 runqueues, but it uses a time-ordered rbtree to build a "timeline" of future
  55 task execution, and thus has no "array switch" artifacts (by which both the
  56 previous vanilla scheduler and RSDL/SD are affected).
  57
  58 CFS also maintains the rq->cfs.min_vruntime value, which is a monotonic
  59 increasing value tracking the smallest vruntime among all tasks in the
  60 runqueue.  The total amount of work done by the system is tracked using
  61 min_vruntime; that value is used to place newly activated entities on the left
  62 side of the tree as much as possible.
  63
  64 The total number of running tasks in the runqueue is accounted through the
  65 rq->cfs.load value, which is the sum of the weights of the tasks queued on the
  66 runqueue.
  67
  68 CFS maintains a time-ordered rbtree, where all runnable tasks are sorted by the
  69 p->se.vruntime key. CFS picks the "leftmost" task from this tree and sticks to it.
  70 As the system progresses forwards, the executed tasks are put into the tree
  71 more and more to the right --- slowly but surely giving a chance for every task
  72 to become the "leftmost task" and thus get on the CPU within a deterministic
  73 amount of time.
  74
  75 Summing up, CFS works like this: it runs a task a bit, and when the task
  76 schedules (or a scheduler tick happens) the task's CPU usage is "accounted
  77 for": the (small) time it just spent using the physical CPU is added to
  78 p->se.vruntime.  Once p->se.vruntime gets high enough so that another task
  79 becomes the "leftmost task" of the time-ordered rbtree it maintains (plus a
  80 small amount of "granularity" distance relative to the leftmost task so that we
  81 do not over-schedule tasks and trash the cache), then the new leftmost task is
  82 picked and the current task is preempted.
  83
  84
  85
  86 4.  SOME FEATURES OF CFS
  87
  88 CFS uses nanosecond granularity accounting and does not rely on any jiffies or
  89 other HZ detail.  Thus the CFS scheduler has no notion of "timeslices" in the
  90 way the previous scheduler had, and has no heuristics whatsoever.  There is
  91 only one central tunable (you have to switch on CONFIG_SCHED_DEBUG):
  92
  93    /proc/sys/kernel/sched_min_granularity_ns
  94
  95 which can be used to tune the scheduler from "desktop" (i.e., low latencies) to
  96 "server" (i.e., good batching) workloads.  It defaults to a setting suitable
  97 for desktop workloads.  SCHED_BATCH is handled by the CFS scheduler module too.
  98
  99 Due to its design, the CFS scheduler is not prone to any of the "attacks" that
 100 exist today against the heuristics of the stock scheduler: fiftyp.c, thud.c,
 101 chew.c, ring-test.c, massive_intr.c all work fine and do not impact
 102 interactivity and produce the expected behavior.
 103
 104 The CFS scheduler has a much stronger handling of nice levels and SCHED_BATCH
 105 than the previous vanilla scheduler: both types of workloads are isolated much
 106 more aggressively.
 107
 108 SMP load-balancing has been reworked/sanitized: the runqueue-walking
 109 assumptions are gone from the load-balancing code now, and iterators of the
 110 scheduling modules are used.  The balancing code got quite a bit simpler as a
 111 result.
 112
 113
 114
 115 5. Scheduling policies
 116
 117 CFS implements three scheduling policies:
 118
 119   - SCHED_NORMAL (traditionally called SCHED_OTHER): The scheduling
 120     policy that is used for regular tasks.
 121
 122   - SCHED_BATCH: Does not preempt nearly as often as regular tasks
 123     would, thereby allowing tasks to run longer and make better use of
 124     caches but at the cost of interactivity. This is well suited for
 125     batch jobs.
 126
 127   - SCHED_IDLE: This is even weaker than nice 19, but its not a true
 128     idle timer scheduler in order to avoid to get into priority
 129     inversion problems which would deadlock the machine.
 130
 131 SCHED_FIFO/_RR are implemented in sched/rt.c and are as specified by
 132 POSIX.
 133
 134 The command chrt from util-linux-ng 2.13.1.1 can set all of these except
 135 SCHED_IDLE.
 136
 137
 138
 139 6.  SCHEDULING CLASSES
 140
 141 The new CFS scheduler has been designed in such a way to introduce "Scheduling
 142 Classes," an extensible hierarchy of scheduler modules.  These modules
 143 encapsulate scheduling policy details and are handled by the scheduler core
 144 without the core code assuming too much about them.
 145
 146 sched/fair.c implements the CFS scheduler described above.
 147
 148 sched/rt.c implements SCHED_FIFO and SCHED_RR semantics, in a simpler way than
 149 the previous vanilla scheduler did.  It uses 100 runqueues (for all 100 RT
 150 priority levels, instead of 140 in the previous scheduler) and it needs no
 151 expired array.
 152
 153 Scheduling classes are implemented through the sched_class structure, which
 154 contains hooks to functions that must be called whenever an interesting event
 155 occurs.
 156
 157 This is the (partial) list of the hooks:
 158
 159  - enqueue_task(...)
 160
 161    Called when a task enters a runnable state.
 162    It puts the scheduling entity (task) into the red-black tree and
 163    increments the nr_running variable.
 164
 165  - dequeue_task(...)
 166
 167    When a task is no longer runnable, this function is called to keep the
 168    corresponding scheduling entity out of the red-black tree.  It decrements
 169    the nr_running variable.
 170
 171  - yield_task(...)
 172
 173    This function is basically just a dequeue followed by an enqueue, unless the
 174    compat_yield sysctl is turned on; in that case, it places the scheduling
 175    entity at the right-most end of the red-black tree.
 176
 177  - check_preempt_curr(...)
 178
 179    This function checks if a task that entered the runnable state should
 180    preempt the currently running task.
 181
 182  - pick_next_task(...)
 183
 184    This function chooses the most appropriate task eligible to run next.
 185
 186  - set_curr_task(...)
 187
 188    This function is called when a task changes its scheduling class or changes
 189    its task group.
 190
 191  - task_tick(...)
 192
 193    This function is mostly called from time tick functions; it might lead to
 194    process switch.  This drives the running preemption.
 195
 196
 197
 198
 199 7.  GROUP SCHEDULER EXTENSIONS TO CFS
 200
 201 Normally, the scheduler operates on individual tasks and strives to provide
 202 fair CPU time to each task.  Sometimes, it may be desirable to group tasks and
 203 provide fair CPU time to each such task group.  For example, it may be
 204 desirable to first provide fair CPU time to each user on the system and then to
 205 each task belonging to a user.
 206
 207 CONFIG_CGROUP_SCHED strives to achieve exactly that.  It lets tasks to be
 208 grouped and divides CPU time fairly among such groups.
 209
 210 CONFIG_RT_GROUP_SCHED permits to group real-time (i.e., SCHED_FIFO and
 211 SCHED_RR) tasks.
 212
 213 CONFIG_FAIR_GROUP_SCHED permits to group CFS (i.e., SCHED_NORMAL and
 214 SCHED_BATCH) tasks.
 215
 216    These options need CONFIG_CGROUPS to be defined, and let the administrator
 217    create arbitrary groups of tasks, using the "cgroup" pseudo filesystem.  See
 218    Documentation/cgroup-v1/cgroups.txt for more information about this filesystem.
 219
 220 When CONFIG_FAIR_GROUP_SCHED is defined, a "cpu.shares" file is created for each
 221 group created using the pseudo filesystem.  See example steps below to create
 222 task groups and modify their CPU share using the "cgroups" pseudo filesystem.
 223
 224         # mount -t tmpfs cgroup_root /sys/fs/cgroup
 225         # mkdir /sys/fs/cgroup/cpu
 226         # mount -t cgroup -ocpu none /sys/fs/cgroup/cpu
 227         # cd /sys/fs/cgroup/cpu
 228
 229         # mkdir multimedia      # create "multimedia" group of tasks
 230         # mkdir browser         # create "browser" group of tasks
 231
 232         # #Configure the multimedia group to receive twice the CPU bandwidth
 233         # #that of browser group
 234
 235         # echo 2048 > multimedia/cpu.shares
 236         # echo 1024 > browser/cpu.shares
 237
 238         # firefox &     # Launch firefox and move it to "browser" group
 239         # echo <firefox_pid> > browser/tasks
 240
 241         # #Launch gmplayer (or your favourite movie player)
 242         # echo <movie_player_pid> > multimedia/tasks