A performability model for soft real-time systems

Presents a simple model that combines the failure to meet deadlines with hardware/software failures in computing the reliability of a real-time system. We define the performability as the probability of meeting deadlines by real-time tasks in the presence of hardware and software failures. Deadline-driven schedules rely on worst-case task execution time. This may be necessary for hard real-time systems, where missed deadlines can be very costly. For soft real-time systems (where a missed deadline is not catastrophic), using worst-case task execution times leads to very inefficient use of processing resources. This is particularly true when worst-case execution occurs very infrequently. Our performability analysis permits task schedules to slide (i.e. require more time than predicted). The amount of slack allowed by the task deadlines can be varied to achieve a desired performability. We are developing a tool that empirically calculates the performability for a task system with specified task profiles and reliability system components.<>