An empirical analysis of algorithms for partially Clairvoyant scheduling

Abstract

We contrast the performance of three algorithms for the problem of deciding whether a Partially Clairvoyant real-time system with relative timing constraints, as specified in the E-T-C scheduling framework, has a feasible schedule. In the E-T-C scheduling model, real-time scheduling problems are specified through a specialized class of constraint logic programs (CLPs) called Quantified Linear Programs (QLPs) [Subramani, K., 2003, An analysis of quantified linear programs. In: C.S. Calude (Ed.) Proceedings of the 4th International Conference on Discrete Mathematics and Theoretical Computer Science (DMTCS), volume 2731 of Lecture Notes in Computer Science, July (Springer-Verlag), pp. 265– 277]; thus algorithms for determining the schedulability of instances are procedures to determine the satisfiability of CLPs. Two of these algorithms, viz., the primal algorithm and the dual algorithm have already been discussed in the literature, while a third algorithm called the randomized dual algorithm has been recently proposed [Subramani, K. and Desovski, D. 2005, A new verification procedure for partially Clairvoyant scheduling. Proceedings of the 3rd International Conference on Formal Modelling and Analysis of Timed Systems (FORMATS), October; Subramani, K. and Desovski, D., 2005, Out of order quantifier elimination for standard quantified linear programs, Journal of Symbolic Computation, 40, 1383–1396]. Our experiments demonstrate that the dual-based algorithms (i.e. the dual and the randomized dual) are more effective from an implementational perspective; this is surprising since all three algorithms have the same worst case asymptotic complexity.