Management of periodic operations in a real-time computation system

Abstract

To understand the problems of real-time control systems it is beneficial to review briefly some of the characteristics of command and control or commercial multiprogrammed systems. Typically, these systems are not time critical and as a result can schedule programs on a queue basis. The System Executive's master scheduling program keeps track of operational programs ready to run (and also keeps track of free processors if the system is a multiprocessor) and then simply assigns programs running time on a priority basis. If the processor (or processors) happens to be busy at a particular moment, the ready-to-run program list becomes a queue. Thus the timing relationships among operational programs of lesser priority are somewhat random, and there is a somewhat unpredictable wait for a given program. In many applications this random wait is acceptable; however, in many real time control systems, in particular real time avionics and space systems, this randomness is not acceptable. For example, if it is time to execute a program with a precise periodicity requirement there must be a guarantee that the processor (or a processor) is available or has lower-priority, interruptable programs in execution. If all the processors happen to be engaged in executing other programs with precise periodicity requirements at this time, a system bottleneck would exist. This bottleneck would introduce errors in the accuracies of the computations. For example, in an avionics system, if the periodicity is not precisely held, a weapon delivery program could easily cause the weapons to miss the target, or an automatic terrain-following program could cause a plane to crash or be forced to pull up to higher altitudes.