Scheduling multithreaded computations by work stealing

Abstract

This paper studies the problem of efficiently schedulling 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 to execute a fully strict computation on P processors using our work-stealing scheduler is T 1 / P + O ( T ∞ , where T 1 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 required by the execution is at most S 1 P , where S 1 is the minimum serial space requirement. We also show that the expected total communication of the algorithm is at most O ( PT ∞ ( 1 + n d ) S max ), where S max is the size of the largest activation record of any thread and n d is the maximum number of times that any thread synchronizes with its parent. This communication bound justifies 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.

Message-Passing Concurrency for Scalable, Stateful, Reconfigurable Middleware

“…In this section we discuss some of the implementation details related to the runtime system, component scheduling, Workers may run out of ready components to execute, in which case they engage in work stealing [10]. Work stealing involves a thief, a worker with no ready components contacting a victim, the worker with the highest number of ready components, and stealing a batch of half of its ready components.…”

Hierarchical Work-Stealing

“…The performance of the work-stealing algorithm depends heavily on these choices. The theoretical bounds of Blumofe and Leiserson [2] are valid for particular choices. How does a processor choose the next task in its stack?…”

“…The workstealing algorithm [2] is a distributed version of list scheduling achieving a good load-balancing on distributed platforms. However, it may suffer from high communication costs for applications transferring large amounts of data or platforms with complex network topologies.…”

Multicore Scheduling for Lightweight Communicating Processes

“…Despite this, efficient lock-free algorithms are more scalable than their lock-based counterparts [18]. Hence our decision to refine existing wait-free work-stealing for using in our scheduler [12,14]. Figure 2 shows the relationship of the migration window to the run queue.…”