Supporting real-time concurrency

摘要

Concurrent real-time applications are complicated since both timing and consistency constraints must be met for correct performance. Furthermore, techniques to enforce these two forms of constraints are often incompatible. For instance, priority-driven preemptive scheduling, which is optimal for meeting timing constraints in some systems, may leave a shared resource's state inconsistent. On the other hand, mutual exclusion techniques that ensure the consistency of shared resources are not well-suited to meeting timing constraints. This dissertation develops concepts and programming language constructs for facilitating the enforcement of both real-time and consistency constraints in applications with concurrency. Our programming paradigm combines an object-based paradigm for the specification of shared resources, and a distributed transaction-based paradigm for the specification of application processes. Resources provide abstract views of shared system entities, such as devices and data structures. Each resource has a state and defines a set of actions that can be invoked by processes to examine or change its state. A resource also specifies scheduling constraints on the execution of its actions to ensure the maintenance of its state's consistency. Processes access resources by invoking actions and express precedence, consistency and timing constraints on action invocations. The implementation of our language constructs with real-time scheduling and locking for concurrency control is also described, including a novel deadlock prevention technique. The utility of the constructs are demonstrated in two ways. First, we describe their use to solve a general concurrent real-time problem called timed atomic commitment. We then describe how they were used to program a graphic simulation of two robot arms coordinating to pick up a moving object under timing constraints.