Abstract: In this paper, we study the problem of efficiently scheduling a wide class of multithreaded computations, called strict; that is, com- putations in which all dependencies from a thread go to the thread’s ancestors in the computation tree. We present the first scheduling al- gorithm which applies to any strict multithreaded computation and is provably efficient in terms of execution time, space complexity and com- munication 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.
Loading