lk/linux-kernel-test
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

sched: new documentation about CFS

Browse Source 
Rewrite of the CFS documentation - because the old one was sorely
out-dated.

Signed-off-by: Claudio Scordino <claudio@evidence.eu.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This commit is contained in:
Claudio Scordino
2008-08-20 15:18:45 +02:00
committed by Ingo Molnar
parent 3fb669dd6e
commit f58e2c33ff
1 changed files with 214 additions and 149 deletions
Download Patch File Download Diff File Expand all files Collapse all files

319
Documentation/scheduler/sched-design-CFS.txt
View File

@@ -1,151 +1,218 @@
=============
CFS Scheduler
=============
This is the CFS scheduler.
80% of CFS's design can be summed up in a single sentence: CFS basically 1. OVERVIEW
models an "ideal, precise multi-tasking CPU" on real hardware.
"Ideal multi-tasking CPU" is a (non-existent :-)) CPU that has 100% CFS stands for "Completely Fair Scheduler," and is the new "desktop" process
physical power and which can run each task at precise equal speed, in scheduler implemented by Ingo Molnar and merged in Linux 2.6.23. It is the
parallel, each at 1/nr_running speed. For example: if there are 2 tasks replacement for the previous vanilla scheduler's SCHED_OTHER interactivity
running then it runs each at 50% physical power - totally in parallel. code.
On real hardware, we can run only a single task at once, so while that 80% of CFS's design can be summed up in a single sentence: CFS basically models
one task runs, the other tasks that are waiting for the CPU are at a an "ideal, precise multi-tasking CPU" on real hardware.
disadvantage - the current task gets an unfair amount of CPU time. In
CFS this fairness imbalance is expressed and tracked via the per-task
p->wait_runtime (nanosec-unit) value. "wait_runtime" is the amount of
time the task should now run on the CPU for it to become completely fair
and balanced.
( small detail: on 'ideal' hardware, the p->wait_runtime value would "Ideal multi-tasking CPU" is a (non-existent :-)) CPU that has 100% physical
always be zero - no task would ever get 'out of balance' from the power and which can run each task at precise equal speed, in parallel, each at
'ideal' share of CPU time. ) 1/nr_running speed. For example: if there are 2 tasks running, then it runs
each at 50% physical power --- i.e., actually in parallel.
CFS's task picking logic is based on this p->wait_runtime value and it On real hardware, we can run only a single task at once, so we have to
is thus very simple: it always tries to run the task with the largest introduce the concept of "virtual runtime." The virtual runtime of a task
p->wait_runtime value. In other words, CFS tries to run the task with specifies when its next timeslice would start execution on the ideal
the 'gravest need' for more CPU time. So CFS always tries to split up multi-tasking CPU described above. In practice, the virtual runtime of a task
CPU time between runnable tasks as close to 'ideal multitasking is its actual runtime normalized to the total number of running tasks.
hardware' as possible.
Most of the rest of CFS's design just falls out of this really simple
concept, with a few add-on embellishments like nice levels,
multiprocessing and various algorithm variants to recognize sleepers.
In practice it works like this: the system runs a task a bit, and when
the task schedules (or a scheduler tick happens) the task's CPU usage is
'accounted for': the (small) time it just spent using the physical CPU
is deducted from p->wait_runtime. [minus the 'fair share' it would have
gotten anyway]. Once p->wait_runtime gets low enough so that another
task becomes the 'leftmost task' of the time-ordered rbtree it maintains
(plus a small amount of 'granularity' distance relative to the leftmost
task so that we do not over-schedule tasks and trash the cache) then the
new leftmost task is picked and the current task is preempted.
The rq->fair_clock value tracks the 'CPU time a runnable task would have 2. FEW IMPLEMENTATION DETAILS
fairly gotten, had it been runnable during that time'. So by using
rq->fair_clock values we can accurately timestamp and measure the
'expected CPU time' a task should have gotten. All runnable tasks are
sorted in the rbtree by the "rq->fair_clock - p->wait_runtime" key, and
CFS picks the 'leftmost' task and sticks to it. As the system progresses
forwards, newly woken tasks are put into the tree more and more to the
right - slowly but surely giving a chance for every task to become the
'leftmost task' and thus get on the CPU within a deterministic amount of
time.
Some implementation details: In CFS the virtual runtime is expressed and tracked via the per-task
p->se.vruntime (nanosec-unit) value. This way, it's possible to accurately
timestamp and measure the "expected CPU time" a task should have gotten.
- the introduction of Scheduling Classes: an extensible hierarchy of [ small detail: on "ideal" hardware, at any time all tasks would have the same
scheduler modules. These modules encapsulate scheduling policy p->se.vruntime value --- i.e., tasks would execute simultaneously and no task
details and are handled by the scheduler core without the core would ever get "out of balance" from the "ideal" share of CPU time. ]
code assuming about them too much.
- sched_fair.c implements the 'CFS desktop scheduler': it is a CFS's task picking logic is based on this p->se.vruntime value and it is thus
replacement for the vanilla scheduler's SCHED_OTHER interactivity very simple: it always tries to run the task with the smallest p->se.vruntime
code. value (i.e., the task which executed least so far). CFS always tries to split
up CPU time between runnable tasks as close to "ideal multitasking hardware" as
possible.
I'd like to give credit to Con Kolivas for the general approach here: Most of the rest of CFS's design just falls out of this really simple concept,
he has proven via RSDL/SD that 'fair scheduling' is possible and that with a few add-on embellishments like nice levels, multiprocessing and various
it results in better desktop scheduling. Kudos Con! algorithm variants to recognize sleepers.
The CFS patch uses a completely different approach and implementation
from RSDL/SD. My goal was to make CFS's interactivity quality exceed
that of RSDL/SD, which is a high standard to meet :-) Testing
feedback is welcome to decide this one way or another. [ and, in any
case, all of SD's logic could be added via a kernel/sched_sd.c module
as well, if Con is interested in such an approach. ]
CFS's design is quite radical: it does not use runqueues, it uses a
time-ordered rbtree to build a 'timeline' of future task execution,
and thus has no 'array switch' artifacts (by which both the vanilla
scheduler and RSDL/SD are affected).
CFS uses nanosecond granularity accounting and does not rely on any 3. THE RBTREE
jiffies or other HZ detail. Thus the CFS scheduler has no notion of
'timeslices' and has no heuristics whatsoever. There is only one CFS's design is quite radical: it does not use the old data structures for the
central tunable (you have to switch on CONFIG_SCHED_DEBUG): runqueues, but it uses a time-ordered rbtree to build a "timeline" of future
task execution, and thus has no "array switch" artifacts (by which both the
previous vanilla scheduler and RSDL/SD are affected).
CFS also maintains the rq->cfs.min_vruntime value, which is a monotonic
increasing value tracking the smallest vruntime among all tasks in the
runqueue. The total amount of work done by the system is tracked using
min_vruntime; that value is used to place newly activated entities on the left
side of the tree as much as possible.
The total number of running tasks in the runqueue is accounted through the
rq->cfs.load value, which is the sum of the weights of the tasks queued on the
runqueue.
CFS maintains a time-ordered rbtree, where all runnable tasks are sorted by the
p->se.vruntime key (there is a subtraction using rq->cfs.min_vruntime to
account for possible wraparounds). CFS picks the "leftmost" task from this
tree and sticks to it.
As the system progresses forwards, the executed tasks are put into the tree
more and more to the right --- slowly but surely giving a chance for every task
to become the "leftmost task" and thus get on the CPU within a deterministic
amount of time.
Summing up, CFS works like this: it runs a task a bit, and when the task
schedules (or a scheduler tick happens) the task's CPU usage is "accounted
for": the (small) time it just spent using the physical CPU is added to
p->se.vruntime. Once p->se.vruntime gets high enough so that another task
becomes the "leftmost task" of the time-ordered rbtree it maintains (plus a
small amount of "granularity" distance relative to the leftmost task so that we
do not over-schedule tasks and trash the cache), then the new leftmost task is
picked and the current task is preempted.
4. SOME FEATURES OF CFS
CFS uses nanosecond granularity accounting and does not rely on any jiffies or
other HZ detail. Thus the CFS scheduler has no notion of "timeslices" in the
way the previous scheduler had, and has no heuristics whatsoever. There is
only one central tunable (you have to switch on CONFIG_SCHED_DEBUG):
/proc/sys/kernel/sched_granularity_ns /proc/sys/kernel/sched_granularity_ns
which can be used to tune the scheduler from 'desktop' (low which can be used to tune the scheduler from "desktop" (i.e., low latencies) to
latencies) to 'server' (good batching) workloads. It defaults to a "server" (i.e., good batching) workloads. It defaults to a setting suitable
setting suitable for desktop workloads. SCHED_BATCH is handled by the for desktop workloads. SCHED_BATCH is handled by the CFS scheduler module too.
CFS scheduler module too.
Due to its design, the CFS scheduler is not prone to any of the Due to its design, the CFS scheduler is not prone to any of the "attacks" that
'attacks' that exist today against the heuristics of the stock exist today against the heuristics of the stock scheduler: fiftyp.c, thud.c,
scheduler: fiftyp.c, thud.c, chew.c, ring-test.c, massive_intr.c all chew.c, ring-test.c, massive_intr.c all work fine and do not impact
work fine and do not impact interactivity and produce the expected interactivity and produce the expected behavior.
behavior.
the CFS scheduler has a much stronger handling of nice levels and The CFS scheduler has a much stronger handling of nice levels and SCHED_BATCH
SCHED_BATCH: both types of workloads should be isolated much more than the previous vanilla scheduler: both types of workloads are isolated much
agressively than under the vanilla scheduler. more aggressively.
( another detail: due to nanosec accounting and timeline sorting, SMP load-balancing has been reworked/sanitized: the runqueue-walking
sched_yield() support is very simple under CFS, and in fact under assumptions are gone from the load-balancing code now, and iterators of the
CFS sched_yield() behaves much better than under any other scheduling modules are used. The balancing code got quite a bit simpler as a
scheduler i have tested so far. ) result.
- sched_rt.c implements SCHED_FIFO and SCHED_RR semantics, in a simpler
way than the vanilla scheduler does. It uses 100 runqueues (for all
100 RT priority levels, instead of 140 in the vanilla scheduler)
and it needs no expired array.
- reworked/sanitized SMP load-balancing: the runqueue-walking
assumptions are gone from the load-balancing code now, and
iterators of the scheduling modules are used. The balancing code got
quite a bit simpler as a result.
Group scheduler extension to CFS
================================
Normally the scheduler operates on individual tasks and strives to provide 5. SCHEDULING CLASSES
fair CPU time to each task. Sometimes, it may be desirable to group tasks
and provide fair CPU time to each such task group. For example, it may
be desirable to first provide fair CPU time to each user on the system
and then to each task belonging to a user.
CONFIG_FAIR_GROUP_SCHED strives to achieve exactly that. It lets The new CFS scheduler has been designed in such a way to introduce "Scheduling
SCHED_NORMAL/BATCH tasks be be grouped and divides CPU time fairly among such Classes," an extensible hierarchy of scheduler modules. These modules
groups. At present, there are two (mutually exclusive) mechanisms to group encapsulate scheduling policy details and are handled by the scheduler core
tasks for CPU bandwidth control purpose: without the core code assuming too much about them.
- Based on user id (CONFIG_FAIR_USER_SCHED) sched_fair.c implements the CFS scheduler described above.
In this option, tasks are grouped according to their user id.
- Based on "cgroup" pseudo filesystem (CONFIG_FAIR_CGROUP_SCHED) sched_rt.c implements SCHED_FIFO and SCHED_RR semantics, in a simpler way than
This options lets the administrator create arbitrary groups the previous vanilla scheduler did. It uses 100 runqueues (for all 100 RT
of tasks, using the "cgroup" pseudo filesystem. See priority levels, instead of 140 in the previous scheduler) and it needs no
Documentation/cgroups.txt for more information about this expired array.
filesystem.
Scheduling classes are implemented through the sched_class structure, which
contains hooks to functions that must be called whenever an interesting event
occurs.
This is the (partial) list of the hooks:
- enqueue_task(...)
Called when a task enters a runnable state.
It puts the scheduling entity (task) into the red-black tree and
increments the nr_running variable.
- dequeue_tree(...)
When a task is no longer runnable, this function is called to keep the
corresponding scheduling entity out of the red-black tree. It decrements
the nr_running variable.
- yield_task(...)
This function is basically just a dequeue followed by an enqueue, unless the
compat_yield sysctl is turned on; in that case, it places the scheduling
entity at the right-most end of the red-black tree.
- check_preempt_curr(...)
This function checks if a task that entered the runnable state should
preempt the currently running task.
- pick_next_task(...)
This function chooses the most appropriate task eligible to run next.
- set_curr_task(...)
This function is called when a task changes its scheduling class or changes
its task group.
- task_tick(...)
This function is mostly called from time tick functions; it might lead to
process switch. This drives the running preemption.
- task_new(...)
The core scheduler gives the scheduling module an opportunity to manage new
task startup. The CFS scheduling module uses it for group scheduling, while
the scheduling module for a real-time task does not use it.
6. GROUP SCHEDULER EXTENSIONS TO CFS
Normally, the scheduler operates on individual tasks and strives to provide
fair CPU time to each task. Sometimes, it may be desirable to group tasks and
provide fair CPU time to each such task group. For example, it may be
desirable to first provide fair CPU time to each user on the system and then to
each task belonging to a user.
CONFIG_GROUP_SCHED strives to achieve exactly that. It lets tasks to be
grouped and divides CPU time fairly among such groups.
CONFIG_RT_GROUP_SCHED permits to group real-time (i.e., SCHED_FIFO and
SCHED_RR) tasks.
CONFIG_FAIR_GROUP_SCHED permits to group CFS (i.e., SCHED_NORMAL and
SCHED_BATCH) tasks.
At present, there are two (mutually exclusive) mechanisms to group tasks for
CPU bandwidth control purposes:
- Based on user id (CONFIG_USER_SCHED)
With this option, tasks are grouped according to their user id.
- Based on "cgroup" pseudo filesystem (CONFIG_CGROUP_SCHED)
This options needs CONFIG_CGROUPS to be defined, and lets the administrator
create arbitrary groups of tasks, using the "cgroup" pseudo filesystem. See
Documentation/cgroups.txt for more information about this filesystem.
Only one of these options to group tasks can be chosen and not both. Only one of these options to group tasks can be chosen and not both.
Group scheduler tunables: When CONFIG_USER_SCHED is defined, a directory is created in sysfs for each new
user and a "cpu_share" file is added in that directory.
When CONFIG_FAIR_USER_SCHED is defined, a directory is created in sysfs for
each new user and a "cpu_share" file is added in that directory.
# cd /sys/kernel/uids # cd /sys/kernel/uids
# cat 512/cpu_share # Display user 512's CPU share # cat 512/cpu_share # Display user 512's CPU share
@@ -155,16 +222,14 @@ each new user and a "cpu_share" file is added in that directory.
2048 2048
# #
CPU bandwidth between two users are divided in the ratio of their CPU shares. CPU bandwidth between two users is divided in the ratio of their CPU shares.
For ex: if you would like user "root" to get twice the bandwidth of user For example: if you would like user "root" to get twice the bandwidth of user
"guest", then set the cpu_share for both the users such that "root"'s "guest," then set the cpu_share for both the users such that "root"'s cpu_share
cpu_share is twice "guest"'s cpu_share is twice "guest"'s cpu_share.
When CONFIG_CGROUP_SCHED is defined, a "cpu.shares" file is created for each
When CONFIG_FAIR_CGROUP_SCHED is defined, a "cpu.shares" file is created group created using the pseudo filesystem. See example steps below to create
for each group created using the pseudo filesystem. See example steps task groups and modify their CPU share using the "cgroups" pseudo filesystem.
below to create task groups and modify their CPU share using the "cgroups"
pseudo filesystem
# mkdir /dev/cpuctl # mkdir /dev/cpuctl
# mount -t cgroup -ocpu none /dev/cpuctl # mount -t cgroup -ocpu none /dev/cpuctl