path: root/man7/sched.7

.\" Copyright (C) 2014 Michael Kerrisk <mtk.manpages@gmail.com>
.\" and Copyright (C) 2014 Peter Zijlstra <peterz@infradead.org>
.\" and Copyright (C) 2014 Juri Lelli <juri.lelli@gmail.com>
.\" Various pieces from the old sched_setscheduler(2) page
.\" 	Copyright (C) Tom Bjorkholm, Markus Kuhn & David A. Wheeler 1996-1999
.\" 	and Copyright (C) 2007 Carsten Emde <Carsten.Emde@osadl.org>
.\" 	and Copyright (C) 2008 Michael Kerrisk <mtk.manpages@gmail.com>
.\"
.\" SPDX-License-Identifier: GPL-2.0-or-later
.\"
.\" Worth looking at: http://rt.wiki.kernel.org/index.php
.\"
.TH sched 7 2023-02-10 "Linux man-pages 6.05.01"
.SH NAME
sched \- overview of CPU scheduling
.SH DESCRIPTION
Since Linux 2.6.23, the default scheduler is CFS,
the "Completely Fair Scheduler".
The CFS scheduler replaced the earlier "O(1)" scheduler.
.\"
.SS API summary
Linux provides the following system calls for controlling
the CPU scheduling behavior, policy, and priority of processes
(or, more precisely, threads).
.TP
.BR nice (2)
Set a new nice value for the calling thread,
and return the new nice value.
.TP
.BR getpriority (2)
Return the nice value of a thread, a process group,
or the set of threads owned by a specified user.
.TP
.BR setpriority (2)
Set the nice value of a thread, a process group,
or the set of threads owned by a specified user.
.TP
.BR sched_setscheduler (2)
Set the scheduling policy and parameters of a specified thread.
.TP
.BR sched_getscheduler (2)
Return the scheduling policy of a specified thread.
.TP
.BR sched_setparam (2)
Set the scheduling parameters of a specified thread.
.TP
.BR sched_getparam (2)
Fetch the scheduling parameters of a specified thread.
.TP
.BR sched_get_priority_max (2)
Return the maximum priority available in a specified scheduling policy.
.TP
.BR sched_get_priority_min (2)
Return the minimum priority available in a specified scheduling policy.
.TP
.BR sched_rr_get_interval (2)
Fetch the quantum used for threads that are scheduled under
the "round-robin" scheduling policy.
.TP
.BR sched_yield (2)
Cause the caller to relinquish the CPU,
so that some other thread be executed.
.TP
.BR sched_setaffinity (2)
(Linux-specific)
Set the CPU affinity of a specified thread.
.TP
.BR sched_getaffinity (2)
(Linux-specific)
Get the CPU affinity of a specified thread.
.TP
.BR sched_setattr (2)
Set the scheduling policy and parameters of a specified thread.
This (Linux-specific) system call provides a superset of the functionality of
.BR sched_setscheduler (2)
and
.BR sched_setparam (2).
.TP
.BR sched_getattr (2)
Fetch the scheduling policy and parameters of a specified thread.
This (Linux-specific) system call provides a superset of the functionality of
.BR sched_getscheduler (2)
and
.BR sched_getparam (2).
.\"
.SS Scheduling policies
The scheduler is the kernel component that decides which runnable thread
will be executed by the CPU next.
Each thread has an associated scheduling policy and a \fIstatic\fP
scheduling priority,
.IR sched_priority .
The scheduler makes its decisions based on knowledge of the scheduling
policy and static priority of all threads on the system.
.PP
For threads scheduled under one of the normal scheduling policies
(\fBSCHED_OTHER\fP, \fBSCHED_IDLE\fP, \fBSCHED_BATCH\fP),
\fIsched_priority\fP is not used in scheduling
decisions (it must be specified as 0).
.PP
Processes scheduled under one of the real-time policies
(\fBSCHED_FIFO\fP, \fBSCHED_RR\fP) have a
\fIsched_priority\fP value in the range 1 (low) to 99 (high).
(As the numbers imply, real-time threads always have higher priority
than normal threads.)
Note well: POSIX.1 requires an implementation to support only a
minimum 32 distinct priority levels for the real-time policies,
and some systems supply just this minimum.
Portable programs should use
.BR sched_get_priority_min (2)
and
.BR sched_get_priority_max (2)
to find the range of priorities supported for a particular policy.
.PP
Conceptually, the scheduler maintains a list of runnable
threads for each possible \fIsched_priority\fP value.
In order to determine which thread runs next, the scheduler looks for
the nonempty list with the highest static priority and selects the
thread at the head of this list.
.PP
A thread's scheduling policy determines
where it will be inserted into the list of threads
with equal static priority and how it will move inside this list.
.PP
All scheduling is preemptive: if a thread with a higher static
priority becomes ready to run, the currently running thread
will be preempted and
returned to the wait list for its static priority level.
The scheduling policy determines the
ordering only within the list of runnable threads with equal static
priority.
.SS SCHED_FIFO: First in-first out scheduling
\fBSCHED_FIFO\fP can be used only with static priorities higher than
0, which means that when a \fBSCHED_FIFO\fP thread becomes runnable,
it will always immediately preempt any currently running
\fBSCHED_OTHER\fP, \fBSCHED_BATCH\fP, or \fBSCHED_IDLE\fP thread.
\fBSCHED_FIFO\fP is a simple scheduling
algorithm without time slicing.
For threads scheduled under the
\fBSCHED_FIFO\fP policy, the following rules apply:
.IP \[bu] 3
A running \fBSCHED_FIFO\fP thread that has been preempted by another thread of
higher priority will stay at the head of the list for its priority and
will resume execution as soon as all threads of higher priority are
blocked again.
.IP \[bu]
When a blocked \fBSCHED_FIFO\fP thread becomes runnable, it
will be inserted at the end of the list for its priority.
.IP \[bu]
If a call to
.BR sched_setscheduler (2),
.BR sched_setparam (2),
.BR sched_setattr (2),
.BR pthread_setschedparam (3),
or
.BR pthread_setschedprio (3)
changes the priority of the running or runnable
.B SCHED_FIFO
thread identified by
.I pid
the effect on the thread's position in the list depends on
the direction of the change to threads priority:
.RS
.IP (a) 5
If the thread's priority is raised,
it is placed at the end of the list for its new priority.
As a consequence,
it may preempt a currently running thread with the same priority.
.IP (b)
If the thread's priority is unchanged,
its position in the run list is unchanged.
.IP (c)
If the thread's priority is lowered,
it is placed at the front of the list for its new priority.
.RE
.IP
According to POSIX.1-2008,
changes to a thread's priority (or policy) using any mechanism other than
.BR pthread_setschedprio (3)
should result in the thread being placed at the end of
the list for its priority.
.\" In Linux 2.2.x and Linux 2.4.x, the thread is placed at the front of the queue
.\" In Linux 2.0.x, the Right Thing happened: the thread went to the back -- MTK
.IP \[bu]
A thread calling
.BR sched_yield (2)
will be put at the end of the list.
.PP
No other events will move a thread
scheduled under the \fBSCHED_FIFO\fP policy in the wait list of
runnable threads with equal static priority.
.PP
A \fBSCHED_FIFO\fP
thread runs until either it is blocked by an I/O request, it is
preempted by a higher priority thread, or it calls
.BR sched_yield (2).
.SS SCHED_RR: Round-robin scheduling
\fBSCHED_RR\fP is a simple enhancement of \fBSCHED_FIFO\fP.
Everything
described above for \fBSCHED_FIFO\fP also applies to \fBSCHED_RR\fP,
except that each thread is allowed to run only for a maximum time
quantum.
If a \fBSCHED_RR\fP thread has been running for a time
period equal to or longer than the time quantum, it will be put at the
end of the list for its priority.
A \fBSCHED_RR\fP thread that has
been preempted by a higher priority thread and subsequently resumes
execution as a running thread will complete the unexpired portion of
its round-robin time quantum.
The length of the time quantum can be
retrieved using
.BR sched_rr_get_interval (2).
.\" On Linux 2.4, the length of the RR interval is influenced
.\" by the process nice value -- MTK
.\"
.SS SCHED_DEADLINE: Sporadic task model deadline scheduling
Since Linux 3.14, Linux provides a deadline scheduling policy
.RB ( SCHED_DEADLINE ).
This policy is currently implemented using
GEDF (Global Earliest Deadline First)
in conjunction with CBS (Constant Bandwidth Server).
To set and fetch this policy and associated attributes,
one must use the Linux-specific
.BR sched_setattr (2)
and
.BR sched_getattr (2)
system calls.
.PP
A sporadic task is one that has a sequence of jobs, where each
job is activated at most once per period.
Each job also has a
.IR "relative deadline" ,
before which it should finish execution, and a
.IR "computation time" ,
which is the CPU time necessary for executing the job.
The moment when a task wakes up
because a new job has to be executed is called the
.I arrival time
(also referred to as the request time or release time).
The
.I start time
is the time at which a task starts its execution.
The
.I absolute deadline
is thus obtained by adding the relative deadline to the arrival time.
.PP
The following diagram clarifies these terms:
.PP
.in +4n
.EX
arrival/wakeup                    absolute deadline
     |    start time                    |
     |        |                         |
     v        v                         v
-----x--------xooooooooooooooooo--------x--------x---
              |<- comp. time ->|
     |<------- relative deadline ------>|
     |<-------------- period ------------------->|
.EE
.in
.PP
When setting a
.B SCHED_DEADLINE
policy for a thread using
.BR sched_setattr (2),
one can specify three parameters:
.IR Runtime ,
.IR Deadline ,
and
.IR Period .
These parameters do not necessarily correspond to the aforementioned terms:
usual practice is to set Runtime to something bigger than the average
computation time (or worst-case execution time for hard real-time tasks),
Deadline to the relative deadline, and Period to the period of the task.
Thus, for
.B SCHED_DEADLINE
scheduling, we have:
.PP
.in +4n
.EX
arrival/wakeup                    absolute deadline
     |    start time                    |
     |        |                         |
     v        v                         v
-----x--------xooooooooooooooooo--------x--------x---
              |<-- Runtime ------->|
     |<----------- Deadline ----------->|
     |<-------------- Period ------------------->|
.EE
.in
.PP
The three deadline-scheduling parameters correspond to the
.IR sched_runtime ,
.IR sched_deadline ,
and
.I sched_period
fields of the
.I sched_attr
structure; see
.BR sched_setattr (2).
These fields express values in nanoseconds.
.\" FIXME It looks as though specifying sched_period as 0 means
.\" "make sched_period the same as sched_deadline".
.\" This needs to be documented.
If
.I sched_period
is specified as 0, then it is made the same as
.IR sched_deadline .
.PP
The kernel requires that:
.PP
.in +4n
.EX
sched_runtime <= sched_deadline <= sched_period
.EE
.in
.PP
.\" See __checkparam_dl in kernel/sched/core.c
In addition, under the current implementation,
all of the parameter values must be at least 1024
(i.e., just over one microsecond,
which is the resolution of the implementation), and less than 2\[ha]63.
If any of these checks fails,
.BR sched_setattr (2)
fails with the error
.BR EINVAL .
.PP
The CBS guarantees non-interference between tasks, by throttling
threads that attempt to over-run their specified Runtime.
.PP
To ensure deadline scheduling guarantees,
the kernel must prevent situations where the set of
.B SCHED_DEADLINE
threads is not feasible (schedulable) within the given constraints.
The kernel thus performs an admittance test when setting or changing
.B SCHED_DEADLINE
policy and attributes.
This admission test calculates whether the change is feasible;
if it is not,
.BR sched_setattr (2)
fails with the error
.BR EBUSY .
.PP
For example, it is required (but not necessarily sufficient) for
the total utilization to be less than or equal to the total number of
CPUs available, where, since each thread can maximally run for
Runtime per Period, that thread's utilization is its
Runtime divided by its Period.
.PP
In order to fulfill the guarantees that are made when
a thread is admitted to the
.B SCHED_DEADLINE
policy,
.B SCHED_DEADLINE
threads are the highest priority (user controllable) threads in the
system; if any
.B SCHED_DEADLINE
thread is runnable,
it will preempt any thread scheduled under one of the other policies.
.PP
A call to
.BR fork (2)
by a thread scheduled under the
.B SCHED_DEADLINE
policy fails with the error
.BR EAGAIN ,
unless the thread has its reset-on-fork flag set (see below).
.PP
A
.B SCHED_DEADLINE
thread that calls
.BR sched_yield (2)
will yield the current job and wait for a new period to begin.
.\"
.\" FIXME Calling sched_getparam() on a SCHED_DEADLINE thread
.\" fails with EINVAL, but sched_getscheduler() succeeds.
.\" Is that intended? (Why?)
.\"
.SS SCHED_OTHER: Default Linux time-sharing scheduling
\fBSCHED_OTHER\fP can be used at only static priority 0
(i.e., threads under real-time policies always have priority over
.B SCHED_OTHER
processes).
\fBSCHED_OTHER\fP is the standard Linux time-sharing scheduler that is
intended for all threads that do not require the special
real-time mechanisms.
.PP
The thread to run is chosen from the static
priority 0 list based on a \fIdynamic\fP priority that is determined only
inside this list.
The dynamic priority is based on the nice value (see below)
and is increased for each time quantum the thread is ready to run,
but denied to run by the scheduler.
This ensures fair progress among all \fBSCHED_OTHER\fP threads.
.PP
In the Linux kernel source code, the
.B SCHED_OTHER
policy is actually named
.BR SCHED_NORMAL .
.\"
.SS The nice value
The nice value is an attribute
that can be used to influence the CPU scheduler to
favor or disfavor a process in scheduling decisions.
It affects the scheduling of
.B SCHED_OTHER
and
.B SCHED_BATCH
(see below) processes.
The nice value can be modified using
.BR nice (2),
.BR setpriority (2),
or
.BR sched_setattr (2).
.PP
According to POSIX.1, the nice value is a per-process attribute;
that is, the threads in a process should share a nice value.
However, on Linux, the nice value is a per-thread attribute:
different threads in the same process may have different nice values.
.PP
The range of the nice value
varies across UNIX systems.
On modern Linux, the range is \-20 (high priority) to +19 (low priority).
On some other systems, the range is \-20..20.
Very early Linux kernels (before Linux 2.0) had the range \-infinity..15.
.\" Linux before 1.3.36 had \-infinity..15.
.\" Since Linux 1.3.43, Linux has the range \-20..19.
.PP
The degree to which the nice value affects the relative scheduling of
.B SCHED_OTHER
processes likewise varies across UNIX systems and
across Linux kernel versions.
.PP
With the advent of the CFS scheduler in Linux 2.6.23,
Linux adopted an algorithm that causes
relative differences in nice values to have a much stronger effect.
In the current implementation, each unit of difference in the
nice values of two processes results in a factor of 1.25
in the degree to which the scheduler favors the higher priority process.
This causes very low nice values (+19) to truly provide little CPU
to a process whenever there is any other
higher priority load on the system,
and makes high nice values (\-20) deliver most of the CPU to applications
that require it (e.g., some audio applications).
.PP
On Linux, the
.B RLIMIT_NICE
resource limit can be used to define a limit to which
an unprivileged process's nice value can be raised; see
.BR setrlimit (2)
for details.
.PP
For further details on the nice value, see the subsections on
the autogroup feature and group scheduling, below.
.\"
.SS SCHED_BATCH: Scheduling batch processes
(Since Linux 2.6.16.)
\fBSCHED_BATCH\fP can be used only at static priority 0.
This policy is similar to \fBSCHED_OTHER\fP in that it schedules
the thread according to its dynamic priority
(based on the nice value).
The difference is that this policy
will cause the scheduler to always assume
that the thread is CPU-intensive.
Consequently, the scheduler will apply a small scheduling
penalty with respect to wakeup behavior,
so that this thread is mildly disfavored in scheduling decisions.
.PP
.\" The following paragraph is drawn largely from the text that
.\" accompanied Ingo Molnar's patch for the implementation of
.\" SCHED_BATCH.
.\" commit b0a9499c3dd50d333e2aedb7e894873c58da3785
This policy is useful for workloads that are noninteractive,
but do not want to lower their nice value,
and for workloads that want a deterministic scheduling policy without
interactivity causing extra preemptions (between the workload's tasks).
.\"
.SS SCHED_IDLE: Scheduling very low priority jobs
(Since Linux 2.6.23.)
\fBSCHED_IDLE\fP can be used only at static priority 0;
the process nice value has no influence for this policy.
.PP
This policy is intended for running jobs at extremely low
priority (lower even than a +19 nice value with the
.B SCHED_OTHER
or
.B SCHED_BATCH
policies).
.\"
.SS Resetting scheduling policy for child processes
Each thread has a reset-on-fork scheduling flag.
When this flag is set, children created by
.BR fork (2)
do not inherit privileged scheduling policies.
The reset-on-fork flag can be set by either:
.IP \[bu] 3
ORing the
.B SCHED_RESET_ON_FORK
flag into the
.I policy
argument when calling
.BR sched_setscheduler (2)
(since Linux 2.6.32);
or
.IP \[bu]
specifying the
.B SCHED_FLAG_RESET_ON_FORK
flag in
.I attr.sched_flags
when calling
.BR sched_setattr (2).
.PP
Note that the constants used with these two APIs have different names.
The state of the reset-on-fork flag can analogously be retrieved using
.BR sched_getscheduler (2)
and
.BR sched_getattr (2).
.PP
The reset-on-fork feature is intended for media-playback applications,
and can be used to prevent applications evading the
.B RLIMIT_RTTIME
resource limit (see
.BR getrlimit (2))
by creating multiple child processes.
.PP
More precisely, if the reset-on-fork flag is set,
the following rules apply for subsequently created children:
.IP \[bu] 3
If the calling thread has a scheduling policy of
.B SCHED_FIFO
or
.BR SCHED_RR ,
the policy is reset to
.B SCHED_OTHER
in child processes.
.IP \[bu]
If the calling process has a negative nice value,
the nice value is reset to zero in child processes.
.PP
After the reset-on-fork flag has been enabled,
it can be reset only if the thread has the
.B CAP_SYS_NICE
capability.
This flag is disabled in child processes created by
.BR fork (2).
.\"
.SS Privileges and resource limits
Before Linux 2.6.12, only privileged
.RB ( CAP_SYS_NICE )
threads can set a nonzero static priority (i.e., set a real-time
scheduling policy).
The only change that an unprivileged thread can make is to set the
.B SCHED_OTHER
policy, and this can be done only if the effective user ID of the caller
matches the real or effective user ID of the target thread
(i.e., the thread specified by
.IR pid )
whose policy is being changed.
.PP
A thread must be privileged
.RB ( CAP_SYS_NICE )
in order to set or modify a
.B SCHED_DEADLINE
policy.
.PP
Since Linux 2.6.12, the
.B RLIMIT_RTPRIO
resource limit defines a ceiling on an unprivileged thread's
static priority for the
.B SCHED_RR
and
.B SCHED_FIFO
policies.
The rules for changing scheduling policy and priority are as follows:
.IP \[bu] 3
If an unprivileged thread has a nonzero
.B RLIMIT_RTPRIO
soft limit, then it can change its scheduling policy and priority,
subject to the restriction that the priority cannot be set to a
value higher than the maximum of its current priority and its
.B RLIMIT_RTPRIO
soft limit.
.IP \[bu]
If the
.B RLIMIT_RTPRIO
soft limit is 0, then the only permitted changes are to lower the priority,
or to switch to a non-real-time policy.
.IP \[bu]
Subject to the same rules,
another unprivileged thread can also make these changes,
as long as the effective user ID of the thread making the change
matches the real or effective user ID of the target thread.
.IP \[bu]
Special rules apply for the
.B SCHED_IDLE
policy.
Before Linux 2.6.39,
an unprivileged thread operating under this policy cannot
change its policy, regardless of the value of its
.B RLIMIT_RTPRIO
resource limit.
Since Linux 2.6.39,
.\" commit c02aa73b1d18e43cfd79c2f193b225e84ca497c8
an unprivileged thread can switch to either the
.B SCHED_BATCH
or the
.B SCHED_OTHER
policy so long as its nice value falls within the range permitted by its
.B RLIMIT_NICE
resource limit (see
.BR getrlimit (2)).
.PP
Privileged
.RB ( CAP_SYS_NICE )
threads ignore the
.B RLIMIT_RTPRIO
limit; as with older kernels,
they can make arbitrary changes to scheduling policy and priority.
See
.BR getrlimit (2)
for further information on
.BR RLIMIT_RTPRIO .
.SS Limiting the CPU usage of real-time and deadline processes
A nonblocking infinite loop in a thread scheduled under the
.BR SCHED_FIFO ,
.BR SCHED_RR ,
or
.B SCHED_DEADLINE
policy can potentially block all other threads from accessing
the CPU forever.
Before Linux 2.6.25, the only way of preventing a runaway real-time
process from freezing the system was to run (at the console)
a shell scheduled under a higher static priority than the tested application.
This allows an emergency kill of tested
real-time applications that do not block or terminate as expected.
.PP
Since Linux 2.6.25, there are other techniques for dealing with runaway
real-time and deadline processes.
One of these is to use the
.B RLIMIT_RTTIME
resource limit to set a ceiling on the CPU time that
a real-time process may consume.
See
.BR getrlimit (2)
for details.
.PP
Since Linux 2.6.25, Linux also provides two
.I /proc
files that can be used to reserve a certain amount of CPU time
to be used by non-real-time processes.
Reserving CPU time in this fashion allows some CPU time to be
allocated to (say) a root shell that can be used to kill a runaway process.
Both of these files specify time values in microseconds:
.TP
.I /proc/sys/kernel/sched_rt_period_us
This file specifies a scheduling period that is equivalent to
100% CPU bandwidth.
The value in this file can range from 1 to
.BR INT_MAX ,
giving an operating range of 1 microsecond to around 35 minutes.
The default value in this file is 1,000,000 (1 second).
.TP
.I /proc/sys/kernel/sched_rt_runtime_us
The value in this file specifies how much of the "period" time
can be used by all real-time and deadline scheduled processes
on the system.
The value in this file can range from \-1 to
.BR INT_MAX \-1.
Specifying \-1 makes the run time the same as the period;
that is, no CPU time is set aside for non-real-time processes
(which was the behavior before Linux 2.6.25).
The default value in this file is 950,000 (0.95 seconds),
meaning that 5% of the CPU time is reserved for processes that
don't run under a real-time or deadline scheduling policy.
.SS Response time
A blocked high priority thread waiting for I/O has a certain
response time before it is scheduled again.
The device driver writer
can greatly reduce this response time by using a "slow interrupt"
interrupt handler.
.\" as described in
.\" .BR request_irq (9).
.SS Miscellaneous
Child processes inherit the scheduling policy and parameters across a
.BR fork (2).
The scheduling policy and parameters are preserved across
.BR execve (2).
.PP
Memory locking is usually needed for real-time processes to avoid
paging delays; this can be done with
.BR mlock (2)
or
.BR mlockall (2).
.\"
.SS The autogroup feature
.\" commit 5091faa449ee0b7d73bc296a93bca9540fc51d0a
Since Linux 2.6.38,
the kernel provides a feature known as autogrouping to improve interactive
desktop performance in the face of multiprocess, CPU-intensive
workloads such as building the Linux kernel with large numbers of
parallel build processes (i.e., the
.BR make (1)
.B \-j
flag).
.PP
This feature operates in conjunction with the
CFS scheduler and requires a kernel that is configured with
.BR CONFIG_SCHED_AUTOGROUP .
On a running system, this feature is enabled or disabled via the file
.IR /proc/sys/kernel/sched_autogroup_enabled ;
a value of 0 disables the feature, while a value of 1 enables it.
The default value in this file is 1, unless the kernel was booted with the
.I noautogroup
parameter.
.PP
A new autogroup is created when a new session is created via
.BR setsid (2);
this happens, for example, when a new terminal window is started.
A new process created by
.BR fork (2)
inherits its parent's autogroup membership.
Thus, all of the processes in a session are members of the same autogroup.
An autogroup is automatically destroyed when the last process
in the group terminates.
.PP
When autogrouping is enabled, all of the members of an autogroup
are placed in the same kernel scheduler "task group".
The CFS scheduler employs an algorithm that equalizes the
distribution of CPU cycles across task groups.
The benefits of this for interactive desktop performance
can be described via the following example.
.PP
Suppose that there are two autogroups competing for the same CPU
(i.e., presume either a single CPU system or the use of
.BR taskset (1)
to confine all the processes to the same CPU on an SMP system).
The first group contains ten CPU-bound processes from
a kernel build started with
.IR "make\~\-j10" .
The other contains a single CPU-bound process: a video player.
The effect of autogrouping is that the two groups will
each receive half of the CPU cycles.
That is, the video player will receive 50% of the CPU cycles,
rather than just 9% of the cycles,
which would likely lead to degraded video playback.
The situation on an SMP system is more complex,
.\" Mike Galbraith, 25 Nov 2016:
.\"     I'd say something more wishy-washy here, like cycles are
.\"     distributed fairly across groups and leave it at that, as your
.\"     detailed example is incorrect due to SMP fairness (which I don't
.\"     like much because [very unlikely] worst case scenario
.\"     renders a box sized group incapable of utilizing more that
.\"     a single CPU total).  For example, if a group of NR_CPUS
.\"     size competes with a singleton, load balancing will try to give
.\"     the singleton a full CPU of its very own.  If groups intersect for
.\"     whatever reason on say my quad lappy, distribution is 80/20 in
.\"     favor of the singleton.
but the general effect is the same:
the scheduler distributes CPU cycles across task groups such that
an autogroup that contains a large number of CPU-bound processes
does not end up hogging CPU cycles at the expense of the other
jobs on the system.
.PP
A process's autogroup (task group) membership can be viewed via the file
.IR /proc/ pid /autogroup :
.PP
.in +4n
.EX
$ \fBcat /proc/1/autogroup\fP
/autogroup\-1 nice 0
.EE
.in
.PP
This file can also be used to modify the CPU bandwidth allocated
to an autogroup.
This is done by writing a number in the "nice" range to the file
to set the autogroup's nice value.
The allowed range is from +19 (low priority) to \-20 (high priority).
(Writing values outside of this range causes
.BR write (2)
to fail with the error
.BR EINVAL .)
.\" FIXME .
.\" Because of a bug introduced in Linux 4.7
.\" (commit 2159197d66770ec01f75c93fb11dc66df81fd45b made changes
.\" that exposed the fact that autogroup didn't call scale_load()),
.\" it happened that *all* values in this range caused a task group
.\" to be further disfavored by the scheduler, with \-20 resulting
.\" in the scheduler mildly disfavoring the task group and +19 greatly
.\" disfavoring it.
.\"
.\" A patch was posted on 23 Nov 2016
.\" ("sched/autogroup: Fix 64bit kernel nice adjustment";
.\" check later to see in which kernel version it lands.
.PP
The autogroup nice setting has the same meaning as the process nice value,
but applies to distribution of CPU cycles to the autogroup as a whole,
based on the relative nice values of other autogroups.
For a process inside an autogroup, the CPU cycles that it receives
will be a product of the autogroup's nice value
(compared to other autogroups)
and the process's nice value
(compared to other processes in the same autogroup.
.PP
The use of the
.BR cgroups (7)
CPU controller to place processes in cgroups other than the
root CPU cgroup overrides the effect of autogrouping.
.PP
The autogroup feature groups only processes scheduled under
non-real-time policies
.RB ( SCHED_OTHER ,
.BR SCHED_BATCH ,
and
.BR SCHED_IDLE ).
It does not group processes scheduled under real-time and
deadline policies.
Those processes are scheduled according to the rules described earlier.
.\"
.SS The nice value and group scheduling
When scheduling non-real-time processes (i.e., those scheduled under the
.BR SCHED_OTHER ,
.BR SCHED_BATCH ,
and
.B SCHED_IDLE
policies), the CFS scheduler employs a technique known as "group scheduling",
if the kernel was configured with the
.B CONFIG_FAIR_GROUP_SCHED
option (which is typical).
.PP
Under group scheduling, threads are scheduled in "task groups".
Task groups have a hierarchical relationship,
rooted under the initial task group on the system,
known as the "root task group".
Task groups are formed in the following circumstances:
.IP \[bu] 3
All of the threads in a CPU cgroup form a task group.
The parent of this task group is the task group of the
corresponding parent cgroup.
.IP \[bu]
If autogrouping is enabled,
then all of the threads that are (implicitly) placed in an autogroup
(i.e., the same session, as created by
.BR setsid (2))
form a task group.
Each new autogroup is thus a separate task group.
The root task group is the parent of all such autogroups.
.IP \[bu]
If autogrouping is enabled, then the root task group consists of
all processes in the root CPU cgroup that were not
otherwise implicitly placed into a new autogroup.
.IP \[bu]
If autogrouping is disabled, then the root task group consists of
all processes in the root CPU cgroup.
.IP \[bu]
If group scheduling was disabled (i.e., the kernel was configured without
.BR CONFIG_FAIR_GROUP_SCHED ),
then all of the processes on the system are notionally placed
in a single task group.
.PP
Under group scheduling,
a thread's nice value has an effect for scheduling decisions
.IR "only relative to other threads in the same task group" .
This has some surprising consequences in terms of the traditional semantics
of the nice value on UNIX systems.
In particular, if autogrouping
is enabled (which is the default in various distributions), then employing
.BR setpriority (2)
or
.BR nice (1)
on a process has an effect only for scheduling relative
to other processes executed in the same session
(typically: the same terminal window).
.PP
Conversely, for two processes that are (for example)
the sole CPU-bound processes in different sessions
(e.g., different terminal windows,
each of whose jobs are tied to different autogroups),
.I modifying the nice value of the process in one of the sessions
.I has no effect
in terms of the scheduler's decisions relative to the
process in the other session.
.\" More succinctly: the nice(1) command is in many cases a no-op since
.\" Linux 2.6.38.
.\"
A possibly useful workaround here is to use a command such as
the following to modify the autogroup nice value for
.I all
of the processes in a terminal session:
.PP
.in +4n
.EX
$ \fBecho 10 > /proc/self/autogroup\fP
.EE
.in
.SS Real-time features in the mainline Linux kernel
.\" FIXME . Probably this text will need some minor tweaking
.\" ask Carsten Emde about this.
Since Linux 2.6.18, Linux is gradually
becoming equipped with real-time capabilities,
most of which are derived from the former
.I realtime\-preempt
patch set.
Until the patches have been completely merged into the
mainline kernel,
they must be installed to achieve the best real-time performance.
These patches are named:
.PP
.in +4n
.EX
patch\-\fIkernelversion\fP\-rt\fIpatchversion\fP
.EE
.in
.PP
and can be downloaded from
.UR http://www.kernel.org\:/pub\:/linux\:/kernel\:/projects\:/rt/
.UE .
.PP
Without the patches and prior to their full inclusion into the mainline
kernel, the kernel configuration offers only the three preemption classes
.BR CONFIG_PREEMPT_NONE ,
.BR CONFIG_PREEMPT_VOLUNTARY ,
and
.B CONFIG_PREEMPT_DESKTOP
which respectively provide no, some, and considerable
reduction of the worst-case scheduling latency.
.PP
With the patches applied or after their full inclusion into the mainline
kernel, the additional configuration item
.B CONFIG_PREEMPT_RT
becomes available.
If this is selected, Linux is transformed into a regular
real-time operating system.
The FIFO and RR scheduling policies are then used to run a thread
with true real-time priority and a minimum worst-case scheduling latency.
.SH NOTES
The
.BR cgroups (7)
CPU controller can be used to limit the CPU consumption of
groups of processes.
.PP
Originally, Standard Linux was intended as a general-purpose operating
system being able to handle background processes, interactive
applications, and less demanding real-time applications (applications that
need to usually meet timing deadlines).
Although the Linux 2.6
allowed for kernel preemption and the newly introduced O(1) scheduler
ensures that the time needed to schedule is fixed and deterministic
irrespective of the number of active tasks, true real-time computing
was not possible up to Linux 2.6.17.
.SH SEE ALSO
.ad l
.nh
.BR chcpu (1),
.BR chrt (1),
.BR lscpu (1),
.BR ps (1),
.BR taskset (1),
.BR top (1),
.BR getpriority (2),
.BR mlock (2),
.BR mlockall (2),
.BR munlock (2),
.BR munlockall (2),
.BR nice (2),
.BR sched_get_priority_max (2),
.BR sched_get_priority_min (2),
.BR sched_getaffinity (2),
.BR sched_getparam (2),
.BR sched_getscheduler (2),
.BR sched_rr_get_interval (2),
.BR sched_setaffinity (2),
.BR sched_setparam (2),
.BR sched_setscheduler (2),
.BR sched_yield (2),
.BR setpriority (2),
.BR pthread_getaffinity_np (3),
.BR pthread_getschedparam (3),
.BR pthread_setaffinity_np (3),
.BR sched_getcpu (3),
.BR capabilities (7),
.BR cpuset (7)
.ad
.PP
.I Programming for the real world \- POSIX.4
by Bill O.\& Gallmeister, O'Reilly & Associates, Inc., ISBN 1-56592-074-0.
.PP
The Linux kernel source files
.IR \%Documentation/\:scheduler/\:sched\-deadline\:.txt ,
.IR \%Documentation/\:scheduler/\:sched\-rt\-group\:.txt ,
.IR \%Documentation/\:scheduler/\:sched\-design\-CFS\:.txt ,
and
.I \%Documentation/\:scheduler/\:sched\-nice\-design\:.txt