Timed Multitasking for Real-Time Embedded Software

Software in real-time embedded systems differs fundamentally from its desktop or Internet counterparts. Embedded computing is not simply computation on small devices. In most control applications, for example, embedded software engages the physical world. It reacts to physical and user-interaction events, performs computation on limited and competing resources, and produces results that further impact the environment. Of necessity, it acquires some properties of the physical world, most particularly, time. Despite the fact that both value and time affect the physical outputs of embedded systems, these two aspects are developed separately in typical embedded software design. The functionality is determined at design time with assumptions such as zero or a fixed non-zero run-time delay. The actual timing properties are determined at run time by a real-time operating system (RTOS). Typically, an RTOS offers as control of these timing properties one number for each task, a priority. Thus, whether a piece of computation can be finished or brought to a quiescent state at a particular time is totally a dynamic phenomenon, and it depends largely on the hardware platform, when the inputs arrive, what other software is running at that time, and the relative priorities. Typically, these factors are out of the control of embedded system designers, and break the timing assumptions that the control algorithms may rely on. In most control applications, this run-time uncertainty is undesirable or even disastrous. Jie Liu Edward A. Lee Palo Alto Research Center 3333 Coyote Hill Rd. Palo Alto, CA 94304 [email protected] Department of EECS University of California, Berkeley Berkeley, CA 94720 [email protected]