A new scheduling algorithm for general strict multithreaded computations

Abstract

In this paper, we study the problem of efficiently scheduling a wide class of multithreaded computations, called strict; that is, computations in which all dependencies from a thread go to the thread’s ancestors in the computation tree. We present the first scheduling algorithm which applies to any strict multithreaded computation and is provably efficient in terms of execution time, space complexity and communication cost. The algorithm is distributed, randomized, works in an asynchronous way and follows the work-stealing paradigm. Our analysis applies for both shared-memory and distributed-memory parallel machines and generalizes the one presented in [5], which applies only to fully strict multithreaded computations; that is, computations in which all dependencies from a thread go to the thread’s parent.