It would be extremely inefficient for a single process to have complete use of the processor from the time of its start-up to the completion of its operations. One reason for this is that most processes must pause very often to wait for input such as data from I/O devices, keyboard input or disk accesses. In the simplest case therefore a large amount of useable CPU time would be wasted due to this blocking. This would result in a large overall time to carry out a number of independent tasks. The processor must also handle asynchronous software or hardware interrupts which may require high priority service for a short period, requiring the currently active process to be displaced from the processor while the interrupt is handled before normal processing can be resumed. Such interrupts may be caused by events such as an input buffer becoming full which, if not serviced in a timely manner could result in an unrecoverable loss of data.

To increase throughput efficiency most modern operating systems implement a method to allow many processes to be available for running at any one time. Their access to the processor is interleaved and, as the speed of modern processors is high compared to slower I/O devices, it is possible to provide a pseudo real-time response to an active user of the system for any particular process when in reality many processes are being run. The total time to complete a number of tasks will also be less due to less time being wasted waiting for external inputs. Interrupt handling is facilitated as well as this round robin scheduling of normal processes.

This procedure is known as Multitasking (or Multiprogramming) and its sequencing is controlled by an operating system service called the scheduler. Some operating systems (e.g. Windows 3.1) just rely on the process itself giving up the processor at regular intervals or when waiting on I/O. This approach is known as co-operative multitasking and it can have many problems as, if the process does not give up the CPU at the expected time failure can occur. In this case all other processes will be blocked and there will be no way for the operating system to gain control of the processor, most likely resulting in a system lockup or crash.

A better approach, and that used in most operating systems designed for efficient multitasking is that of pre-emptive multitasking . In this case it is the operating system that decides which process gets access to the CPU. It can allocate CPU time to a process or remove the process from the CPU as required. Each candidate process gets access to the CPU for a short time (known as a quantum) based on their allocated priority, their process class and also their voluntary release of the processor as they wait for external input.