Wait-Free Snapshots in Real-Time Systems: Algorithms and Performance

Snap-shot mechanisms are used to read a globally consistent set of variable values. Such a mechanism can be used to solve a variety of communication and synchronization problems, including system monitoring and control of real-time applications. Methods based on locking (e.g. using semaphores) are penalized by blocking, which typically leads to difficulties in guaranteeing deadlines of high priority tasks. Lock-free methods, which take a snap-shot and then check if it corresponds to a consistent system state, have unpredictable timing-behavior, since they may have to retry an unpredictable number of times. Clearly, a method which combines the predictability of locking-based methods with the low interference (no blocking) of lock-free methods is desirable. In this paper we present one such method, based and the concept of wait-freeness. A wait-free method is a lockfree method which is guaranteed to correctly complete in a bounded number of steps. The price to pay for this predictability in the timing domain is the need for more than one copy of the shared objects. In addition to proving our method correct, we evaluate it analytically by formulating and comparing schedulability equations for snapshots in systems using lock-based, lock-free, and our wait-free method. We also outline ideas to study the effects of using the different snapshot methods in distributed (CAN-based) systems. Our evaluations indicate that our method is an efficient and safe alternative to traditional lock-based and lock-free methods. This work is performed within the Advanced Software Technology (ASTEC) competence center, supported by the Swedish board for technical development (NUTEK) Partially supported by a grant of the Swedish Research Council for Engineering Sciences (TFR) zPart of this work has been done while this author has been a visiting researcher at the Real-Time Systems Group, Department of Computer Systems (DOCS), Uppsala University, Sweden