The theory behind task scheduling is very simple. If there are runnable processes in a system, at least one process must always be running. If there are more runnable processes than processors in a system, not all the processes can be running all the time.

Therefore, some processes need to be stopped temporarily, or suspended, so that others can be running again. The scheduler decides what process in the queue will run next.

As already mentioned, Linux, like all other Unix variants, is a multitasking operating system. That means that several tasks can be running at the same time. Linux provides a so called preemptive multitasking, where the scheduler decides when a process is suspended. This forced suspension is called preemption. All Unix flavors have been providing preemptive multitasking since the beginning.

The time period for which a process will be running before it is preempted is defined in advance. It is called a timeslice of a process and represents the amount of processor time that is provided to each process. By assigning timeslices, the scheduler makes global decisions for the running system, and prevents individual processes from dominating over the processor resources.

The scheduler evaluates processes based on their priority. To calculate the current priority of a process, the task scheduler uses complex algorithms. As a result, each process is given a value according to which it is “allowed” to run on a processor.

CFS tries to guarantee a fair approach to each runnable task. To find the most balanced way of task scheduling, it uses the concept of red-black tree. A red-black tree is a type of self-balancing data search tree which provides inserting and removing entries in a reasonable way so that it remains well balanced. For more information, see the wiki pages of Red-black tree.

When a task enters into the run queue (a planned time line of processes to be executed next), the scheduler records the current time. While the process waits for processor time, its “wait” value gets incremented by an amount derived from the total number of tasks currently in the run queue and the process priority. As soon as the processor runs the task, its “wait” value gets decremented. If the value drops below a certain level, the task is preempted by the scheduler and other tasks get closer to the processor. By this algorithm, CFS tries to reach the ideal state where the “wait” value is always zero.

Since the Linux kernel version 2.6.24, CFS can be tuned to be fair to users or groups rather than to tasks only. Runnable tasks are then grouped to form entities, and CFS tries to be fair to these entities instead of individual runnable tasks. The scheduler also tries to be fair to individual tasks within these entities.

Tasks can be grouped in two mutually exclusive ways:

The way the kernel scheduler lets you group the runnable tasks depends on setting the kernel compile-time options CONFIG_FAIR_USER_SCHED and CONFIG_FAIR_CGROUP_SCHED. The default setting in openSUSE® Leap 42.2 is to use control groups, which lets you create groups as needed. For more information, see Chapter 9, Kernel Control Groups.

Basic aspects of the task scheduler behavior can be set through the kernel configuration options. Setting these options is part of the kernel compilation process. Because kernel compilation process is a complex task and out of this document's scope, refer to relevant source of information.

Warning: Kernel Compilation

If you run openSUSE Leap on a kernel that was not shipped with it, for example on a self-compiled kernel, you lose the entire support entitlement.

Documents regarding task scheduling policy often use several technical terms which you need to know to understand the information correctly. Here are some:

The chrt command sets or retrieves the real-time scheduling attributes of a running process, or runs a command with the specified attributes. You can get or retrieve both the scheduling policy and priority of a process.

In the following examples, a process whose PID is 16244 is used.

To retrieve the real-time attributes of an existing task:

root # chrt -p 16244
pid 16244's current scheduling policy: SCHED_OTHER
pid 16244's current scheduling priority: 0

Before setting a new scheduling policy on the process, you need to find out the minimum and maximum valid priorities for each scheduling algorithm:

root # chrt -m
SCHED_OTHER min/max priority : 0/0
SCHED_FIFO min/max priority : 1/99
SCHED_RR min/max priority : 1/99
SCHED_BATCH min/max priority : 0/0
SCHED_IDLE min/max priority : 0/0

In the above example, SCHED_OTHER, SCHED_BATCH, SCHED_IDLE polices only allow for priority 0, while that of SCHED_FIFO and SCHED_RR can range from 1 to 99.

To set SCHED_BATCH scheduling policy:

root # chrt -b -p 0 16244
pid 16244's current scheduling policy: SCHED_BATCH
pid 16244's current scheduling priority: 0

For more information on chrt, see its man page (man 1 chrt).

The sysctl interface for examining and changing kernel parameters at runtime introduces important variables by means of which you can change the default behavior of the task scheduler. The syntax of the sysctl is simple, and all the following commands must be entered on the command line as root.

To read a value from a kernel variable, enter

sysctl variable

To assign a value, enter

sysctl variable=value

To get a list of all scheduler related sysctl variables, enter

sysctl -A | grep "sched" | grep -v"domain"

root # sysctl -A | grep "sched" | grep -v "domain"
kernel.sched_cfs_bandwidth_slice_us = 5000
kernel.sched_child_runs_first = 0
kernel.sched_compat_yield = 0
kernel.sched_latency_ns = 6000000
kernel.sched_migration_cost_ns = 500000
kernel.sched_min_granularity_ns = 2000000
kernel.sched_nr_migrate = 32
kernel.sched_rr_timeslice_ms = 25
kernel.sched_rt_period_us = 1000000
kernel.sched_rt_runtime_us = 950000
kernel.sched_shares_window_ns = 10000000
kernel.sched_time_avg_ms = 1000
kernel.sched_tunable_scaling = 1
kernel.sched_wakeup_granularity_ns = 2500000

Note that variables ending with “_ns” and “_us” accept values in nanoseconds and microseconds, respectively.

A list of the most important task scheduler sysctl tuning variables (located at /proc/sys/kernel/) with a short description follows:

CFS comes with a new improved debugging interface, and provides runtime statistics information. Relevant files were added to the /proc file system, which can be examined simply with the cat or less command. A list of the related /proc files follows with their short description: