The heart of managing the processor comes down to two related issues:
Ensuring that each process and application receives enough of the processor's time to function properly.
Using as many processor cycles for real work as is possible.
The basic unit of software that the operating system deals with in scheduling the work done by the processor is either a process or a thread, depending on the operating system.
It's tempting to think of a process as an application, but that gives an incomplete picture of how processes relate to the operating system and hardware. The application you see (word processor or spreadsheet or game) is, indeed, a process, but that application may cause several other processes to begin, for tasks like communications with other devices or other computers. There are also numerous processes that run without giving you direct evidence that they ever exist. For example, Windows XP and UNIX can have dozens of background processes running to handle the network, memory management, disk management, virus checking and so on.
A process, then, is software that performs some action and can be controlled -- by a user, by other applications or by the operating system.
It is processes, rather than applications, that the operating system controls and schedules for execution by the CPU. In a single-tasking system, the schedule is straightforward. The operating system allows the application to begin running, suspending the execution only long enough to deal with interrupts and user input.
Interrupts are special signals sent by hardware or software to the CPU. It's as if some part of the computer suddenly raised its hand to ask for the CPU's attention in a lively meeting. Sometimes the operating system will schedule the priority of processes so that interrupts are masked -- that is, the operating system will ignore the interrupts from some sources so that a particular job can be finished as quickly as possible. There are some interrupts (such as those from error conditions or problems with memory) that are so important that they can't be ignored. These non-maskable interrupts (NMIs) must be dealt with immediately, regardless of the other tasks at hand.
While interrupts add some complication to the execution of processes in a single-tasking system, the job of the operating system becomes much more complicated in a multi-tasking system. Now, the operating system must arrange the execution of applications so that you believe that there are several things happening at once. This is complicated because the CPU can only do one thing at a time. In order to give the appearance of lots of things happening at the same time, the operating system has to switch between different processes thousands of times a second. Here's how it happens:
A process occupies a certain amount of RAM. It also makes use of registers, stacks and queues within the CPU and operating-system memory space.
When two processes are multi-tasking, the operating system allots a certain number of CPU execution cycles to one program.
After that number of cycles, the operating system makes copies of all the registers, stacks and queues used by the processes, and notes the point at which the process paused in its execution.
It then loads all the registers, stacks and queues used by the second process and allows it a certain number of CPU cycles.
When those are complete, it makes copies of all the registers, stacks and queues used by the second program, and loads the first program.
All of the information needed to keep track of a process when switching is kept in a data package called a process control block. The process control block typically contains:
* An ID number that identifies the process
* Pointers to the locations in the program and its data where processing last occurred
* Register contents
* States of various flags and switches
* Pointers to the upper and lower bounds of the memory required for the process
* A list of files opened by the process
* The priority of the process
* The status of all I/O devices needed by the process
Each process has a status associated with it. Many processes consume no CPU time until they get some sort of input. For example, a process might be waiting on a keystroke from the user. While it is waiting for the keystroke, it uses no CPU time. While it is waiting, it is "suspended". When the keystroke arrives, the OS changes its status. When the status of the process changes, from pending to active, for example, or from suspended to running, the information in the process control block must be used like the data in any other program to direct execution of the task-switching portion of the operating system.
This process swapping happens without direct user interference, and each process gets enough CPU cycles to accomplish its task in a reasonable amount of time. Trouble can come, though, if the user tries to have too many processes functioning at the same time. The operating system itself requires some CPU cycles to perform the saving and swapping of all the registers, queues and stacks of the application processes. If enough processes are started, and if the operating system hasn't been carefully designed, the system can begin to use the vast majority of its available CPU cycles to swap between processes rather than run processes. When this happens, it's called thrashing, and it usually requires some sort of direct user intervention to stop processes and bring order back to the system.
One way that operating-system designers reduce the chance of thrashing is by reducing the need for new processes to perform various tasks. Some operating systems allow for a "process-lite," called a thread, that can deal with all the CPU-intensive work of a normal process, but generally does not deal with the various types of I/O and does not establish structures requiring the extensive process control block of a regular process. A process may start many threads or other processes, but a thread cannot start a process.
So far, all the scheduling we've discussed has concerned a single CPU. In a system with two or more CPUs, the operating system must divide the workload among the CPUs, trying to balance the demands of the required processes with the available cycles on the different CPUs. Asymmetric operating systems use one CPU for their own needs and divide application processes among the remaining CPUs. Symmetric operating systems divide themselves among the various CPUs, balancing demand versus CPU availability even when the operating system itself is all that's running.
Even if the operating system is the only software with execution needs, the CPU is not the only resource to be scheduled. Memory management is the next crucial step in making sure that all processes run smoothly.