For threads scheduled under one of the normal scheduling policies (SCHED_OTHER, SCHED_IDLE, SCHED_BATCH), sched_priority is not used in scheduling decisions (it must be specified as 0).
Processes scheduled under one of the real-time policies (SCHED_FIFO, SCHED_RR) have a sched_priority 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 sched_get_priority_min(2) and sched_get_priority_max(2) to find the range of priorities supported for a particular policy.
Conceptually, the scheduler maintains a list of runnable threads for each possible sched_priority 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.
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.
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.
No other events will move a thread scheduled under the SCHED_FIFO policy in the wait list of runnable threads with equal static priority.
A SCHED_FIFO thread runs until either it is blocked by an I/O request, it is preempted by a higher priority thread, or it calls sched_yield(2).
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 relative deadline, before which it should finish execution, and a 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 arrival time (also referred to as the request time or release time). The start time is the time at which a task starts its execution. The absolute deadline is thus obtained by adding the relative deadline to the arrival time.
The following diagram clarifies these terms:
arrival/wakeup absolute deadline
| start time |
| | |
v v v
-----x--------xooooooooooooooooo--------x--------x---
|<- comp. time ->|
|<------- relative deadline ------>|
|<-------------- period ------------------->|
When setting a SCHED_DEADLINE policy for a thread using sched_setattr(2), one can specify three parameters: Runtime, Deadline, and 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 SCHED_DEADLINE scheduling, we have:
arrival/wakeup absolute deadline
| start time |
| | |
v v v
-----x--------xooooooooooooooooo--------x--------x---
|<-- Runtime ------->|
|<----------- Deadline ----------->|
|<-------------- Period ------------------->|
The three deadline-scheduling parameters correspond to the sched_runtime, sched_deadline, and sched_period fields of the sched_attr structure; see sched_setattr(2). These fields express values in nanoseconds. If sched_period is specified as 0, then it is made the same as sched_deadline.
The kernel requires that:
sched_runtime <= sched_deadline <= sched_period
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 2^63. If any of these checks fails, sched_setattr(2) fails with the error EINVAL.
The CBS guarantees non-interference between tasks, by throttling threads that attempt to over-run their specified Runtime.
To ensure deadline scheduling guarantees, the kernel must prevent situations where the set of SCHED_DEADLINE threads is not feasible (schedulable) within the given constraints. The kernel thus performs an admittance test when setting or changing SCHED_DEADLINE policy and attributes. This admission test calculates whether the change is feasible; if it is not, sched_setattr(2) fails with the error EBUSY.
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.
In order to fulfill the guarantees that are made when a thread is admitted to the SCHED_DEADLINE policy, SCHED_DEADLINE threads are the highest priority (user controllable) threads in the system; if any SCHED_DEADLINE thread is runnable, it will preempt any thread scheduled under one of the other policies.
A call to fork(2) by a thread scheduled under the SCHED_DEADLINE policy fails with the error EAGAIN, unless the thread has its reset-on-fork flag set (see below).
A SCHED_DEADLINE thread that calls sched_yield(2) will yield the current job and wait for a new period to begin.
The thread to run is chosen from the static priority 0 list based on a dynamic 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 SCHED_OTHER threads.
In the Linux kernel source code, the SCHED_OTHER policy is actually named SCHED_NORMAL.
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.
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.
The degree to which the nice value affects the relative scheduling of SCHED_OTHER processes likewise varies across UNIX systems and across Linux kernel versions.
With the advent of the CFS scheduler in kernel 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).
On Linux, the RLIMIT_NICE resource limit can be used to define a limit to which an unprivileged process's nice value can be raised; see setrlimit(2) for details.
For further details on the nice value, see the subsections on the autogroup feature and group scheduling, below.
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).
This policy is intended for running jobs at extremely low priority (lower even than a +19 nice value with the SCHED_OTHER or SCHED_BATCH policies).
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 sched_getscheduler(2) and sched_getattr(2).
The reset-on-fork feature is intended for media-playback applications, and can be used to prevent applications evading the RLIMIT_RTTIME resource limit (see getrlimit(2)) by creating multiple child processes.
More precisely, if the reset-on-fork flag is set, the following rules apply for subsequently created children:
After the reset-on-fork flag has been enabled, it can be reset only if the thread has the CAP_SYS_NICE capability. This flag is disabled in child processes created by fork(2).
A thread must be privileged (CAP_SYS_NICE) in order to set or modify a SCHED_DEADLINE policy.
Since Linux 2.6.12, the RLIMIT_RTPRIO resource limit defines a ceiling on an unprivileged thread's static priority for the SCHED_RR and SCHED_FIFO policies. The rules for changing scheduling policy and priority are as follows:
Privileged (CAP_SYS_NICE) threads ignore the RLIMIT_RTPRIO limit; as with older kernels, they can make arbitrary changes to scheduling policy and priority. See getrlimit(2) for further information on RLIMIT_RTPRIO.
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 RLIMIT_RTTIME resource limit to set a ceiling on the CPU time that a real-time process may consume. See getrlimit(2) for details.
Since version 2.6.25, Linux also provides two /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:
Memory locking is usually needed for real-time processes to avoid paging delays; this can be done with mlock(2) or mlockall(2).
This feature operates in conjunction with the CFS scheduler and requires a kernel that is configured with CONFIG_SCHED_AUTOGROUP. On a running system, this feature is enabled or disabled via the file /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 noautogroup parameter.
A new autogroup is created when a new session is created via setsid(2); this happens, for example, when a new terminal window is started. A new process created by 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.
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.
Suppose that there are two autogroups competing for the same CPU (i.e., presume either a single CPU system or the use of 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 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, 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.
A process's autogroup (task group) membership can be viewed via the file /proc/[pid]/autogroup:
$ cat /proc/1/autogroup /autogroup-1 nice 0
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 write(2) to fail with the error EINVAL.)
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.
The use of the cgroups(7) CPU controller to place processes in cgroups other than the root CPU cgroup overrides the effect of autogrouping.
The autogroup feature groups only processes scheduled under non-real-time policies (SCHED_OTHER, SCHED_BATCH, and SCHED_IDLE). It does not group processes scheduled under real-time and deadline policies. Those processes are scheduled according to the rules described earlier.
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:
Under group scheduling, a thread's nice value has an effect for scheduling decisions 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 setpriority(2) or 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).
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), modifying the nice value of the process in one of the sessions has no effect in terms of the scheduler's decisions relative to the process in the other session. A possibly useful workaround here is to use a command such as the following to modify the autogroup nice value for all of the processes in a terminal session:
$ echo 10 > /proc/self/autogroup
patch-kernelversion-rtpatchversion
and can be downloaded from
Without the patches and prior to their full inclusion into the mainline kernel, the kernel configuration offers only the three preemption classes CONFIG_PREEMPT_NONE, CONFIG_PREEMPT_VOLUNTARY, and CONFIG_PREEMPT_DESKTOP which respectively provide no, some, and considerable reduction of the worst-case scheduling latency.
With the patches applied or after their full inclusion into the mainline kernel, the additional configuration item 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.
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 kernel 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 kernel version 2.6.17.
Programming for the real world - POSIX.4 by Bill O. Gallmeister, O'Reilly & Associates, Inc., ISBN 1-56592-074-0.
The Linux kernel source files Documentation/scheduler/sched-deadline.txt, Documentation/scheduler/sched-rt-group.txt, Documentation/scheduler/sched-design-CFS.txt, and Documentation/scheduler/sched-nice-design.txt