A decision support system for the algorithm-architecture mapping problem

This paper presents a decision support system, termed MAPPER, for the mapping of large scale resource allocation algorithms onto parallel computing architectures. The mapping problem is viewed as one of assigning the nodes of a finite, directed, acyclic task graph (representing the logical and data dependencies among the tasks constituting the algorithm) onto the nodes of a finite, undirected processor graph (denoting the parallel computing architecture). The objective is to minimize the completion time of the algorithm such that the redundancy, processor memory, and security constraints are satisfied. Key features of MAPPER are its graphical user interface, an explicit model of task and message queueing, and the incorporation of precedence relationships among the tasks of the algorithm. Four mapping algorithms are embedded in MAPPER. The first algorithm is a two-stage heuristic that determines the order of task execution based on the critical path method(CPM), and then employs a one-step optimization method to determine the task allocation. The second algorithm employs the idea of pair-wise exchange on task execution order to improve the performance of the heuristic algorithm. The third algorithm is an optimal mapping strategy using A* algorithm, wherein a lower bound on the cost-to-go is obtained from the two-stage heuristic. Finally, an /spl epsiv/-optimal algorithm, A*/sub /spl epsiv//, is developed to subdue the combinatorial explosion of the A* algorithm.