A new list scheduling method based on structural properties of task graphs

We propose a new list scheduling method for obtaining a quasi-optimal solution to the multiprocessor scheduling problem that has been recognized as a key problem for realizing parallelizing compilers. Our approach for achieving a nearly optimal performance is to take a linear combination of several characteristics to obtain a heuristic function instead of taking a lexicographical combination. The goodness of the resultant heuristics is evaluated by experiments.