Validation of timing and communication constraints in real-time parallel programs

Abstract

Two notable characteristics of real-time parallel programs are constraints on the timing behaviors and concurrency among tasks. Timing behaviors and concurrency are associated with uncertainty which may incur subtle yet critical timing and communicating errors. In this thesis, we deal with three kinds of uncertainty in real-time parallel programs and their specifications - timing uncertainty, nondeterminacy, and incompleteness. In particular, we describe the implications of the uncertainty over the verification phase and suggest a few ways of testing and verifying real-time parallel programs with uncertainty. Timing aspects of real-time programs are affected by such factors as external interfaces and underlying platforms. Since it is almost impossible to predict and/or measure their timing behaviors precisely, we propose a method of coping with such timing uncertainty in a formal fashion by incorporating fuzzy concepts into the timing model of duration calculus formalism. Semantics and proof systems for fuzzy duration calculus are also defined and illustrated using simple examples. Timing constraints on real-time programs can be verified against the fuzzy duration formulas based on the proof systems. Compared to the classical formalism, fuzzy duration calculus allows us to capture the vagueness of state intervals using fuzzy numbers. To demonstrate the effectiveness of the proposed ideas, a part of nuclear power plant control system is specified and verified using the fuzzy duration calculus. The sequence of interactions among concurrent tasks can change for reasons such as relative progress of tasks and communication delay. Uncertainty in the interaction sequences results in the nondeterminacy of parallel programs. Selective Instrumentation, a reproduction method with minimal probe effects, is suggested in this thesis to handle the difficulties associated with nondeterminacy --- non-reproducibility, probe effects and the space overhead. In order to minimize ...