103.6 Lesson 1

Linux, as a preemptive multi-processing operating system, implements a scheduler that organizes the process queue. More precisely, the scheduler also decides which queued thread will be executed — a process can branch out many independent threads — but process and thread are interchangeable terms in this context. Every process has two predicates that intervene on its scheduling: the scheduling policy and the scheduling priority.

There are two main types of scheduling policies: real-time policies and normal policies. Processes under a real-time policy are scheduled by their priority values directly. If a more important process becomes ready to run, a less important running process is preempted and the higher priority process takes control of the CPU. A lower priority process will gain CPU control only if higher priority processes are idle or waiting for hardware response.

Any real-time process has higher priority than a normal process. As a general purpose operating system, Linux runs just a few real-time processes. Most processes, including system and user programs, run under normal scheduling policies. Normal processes usually have the same priority value, but normal policies can define execution priority rules using another process predicate: the nice value. To avoid confusion with the dynamic priorities derived from nice values, scheduling priorities are usually called static scheduling priorities.

The Linux scheduler can be configured in many different ways and even more intricate ways of establishing priorities exist, but these general concepts always apply. When inspecting and tuning process scheduling, it is important to keep in mind that only processes under normal scheduling policy will be affected.

Linux reserves static priorities ranging from 0 to 99 for real-time processes and normal processes are assigned to static priorities ranging from 100 to 139, meaning that there are 39 different priority levels for normal processes. Lower values mean higher priority. The static priority of an active process can be found in the sched file, located in its respective directory inside the /proc filesystem:

As shown in the example, the line beginning with prio gives the priority value of the process (the PID 1 process is the init or the systemd process, the first process the kernel starts during system initialization). The standard priority for normal processes is 120, so that it can be decreased to 100 or increased to 139. The priorities of all running process can be verified with the command ps -Al or ps -el:

The PRI column indicates the static priority assigned by the kernel. Note, however, that the priority value displayed by ps differs from that obtained in the previous example. Due to historical reasons, priorities displayed by ps range from -40 to 99 by default, so the actual priority is obtained by adding 40 to it (in particular, 80 + 40 = 120).

It is also possible to continuously monitor processes currently being managed by the Linux kernel with program top. As with ps, top also displays the priority value differently. To make it easier to identify real-time processes, top subtracts the priority value by 100, thus making all real-time priorities negative, with a negative number or rt identifying them. Therefore, normal priorities displayed by top range from 0 to 39.

Every normal process begins with a default nice value of 0 (priority 120). The nice name comes from the idea that “nicer” processes allow other processes to run before them in a particular execution queue. Nice numbers range from -20 (less nice, high priority) to 19 (more nice, low priority). Linux also allows the ability to assign different nice values to threads within the same process. The NI column in ps output indicates the nice number.

Only the root user can decrease the niceness of a process below zero. It’s possible to start a process with a non-standard priority with the command nice. By default, nice changes the niceness to 10, but it can be specified with option -n:

In this example, the command tar is executed with a niceness of 15. The command renice can be used to change the priority of a running process. The option -p indicates the PID number of the target process. For example:

The options -g and -u are used to modify all the processes of a specific group or user, respectively. With renice +5 -g users, the niceness of processes owned by users of the group users will be raised in five.

Besides renice, the priority of processes can be modified with other programs, like the program top. On the top main screen, the niceness of a process can be modified by pressing r and then the PID number of the process:

The message PID to renice [default pid = 1] appears with the first listed process selected by default. To change the priority of another process, type its PID and press Enter. Then, the message Renice PID 1 to value will appear (with the requested PID number) and a new nice value can be assigned.