Scheduling multithreaded computations by work stealing

This paper studies the problem of efficiently scheduling fully strict (i.e., well-structured) multithreaded computations on parallel computers. A popular and practical method of scheduling this kind of dynamic MIMD-style computation is “work stealing,” in which processors needing work steal computational threads from other processors. In this paper, we give the first provably good work-stealing scheduler for multithreaded computations with dependencies. Specifically, our analysis shows that the expected time T_P to execute a fully strict computation on P processors using our work-stealing scheduler is T_P=O(T₁/P+T_∞ ), where T₁ is the minimum serial execution time of the multithreaded computation and T_∞ is the minimum execution time with an infinite number of processors. Moreover, the space S_P required by the execution satisfies S_P⩽S₁P. We also show that the expected total communication of the algorithm is at most O(T_∞S_max P), where S_max is the size of the largest activation record of any thread, thereby justifying the folk wisdom that work-stealing schedulers are more communication efficient than their work-sharing counterparts. All three of these bounds are existentially optimal to within a constant factor