Parallelized search for the optimal/sub-optimal solutions of task scheduling problem taking account of communication overhead

In order to achieve the efficient parallel processing on a given multiprocessor system, it is important to draw the performance of the multiprocessor system sufficiently. The task scheduling technique, in which the extracted task set from an application is assigned onto multiple processors to be executed in parallel in the minimum processing time, is very important key for the efficient parallel processing. In this paper, we first propose new priority levels with communication overhead to find better solutions of any scheduling problem in early stage of search. The new priority levels are useful to bound the branches of search tree to be efficiently cut off with better estimated lower estimation of scheduling length. Second, using the new priority levels, our search method performs depth-first search, which is full-search method and has the possibility to got an optimal solutions. In the case of large scale task graphs, full-search is impossible. However, even if the search processes are interrupted in any given time, it can find better solutions in the given time than past algorithms Third, in order to shorten the execution time of search process for the scheduling problem, we show parallel search methods and the results of the efficiency of the parallel search. The parallel search methods are implemented on HITACHI SR8000 multiprocessor system which has 8 processors by using MP.