Scheduling parallel programs by work stealing with private deques

Acar, Umut A.; Charguéraud, Arthur; Rainey, Mike

doi:10.1145/2517327.2442538

Cited by 32 publications

(25 citation statements)

References 37 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…A process, comprising multiple lightweight threads, is mapped to the required number of virtual processors using a provision operation. There is a simple built-in mechanism for basic load-balancing, extended using a low-latency work-stealing mechanism (Acar et al, 2013). Manticore also provides a low-level internal scheduling interface (Acar et al, 2012), which provides preemption and interrupt via a signal-based approach.…”

Section: Manticorementioning

confidence: 99%

PAEAN: Portable and scalable runtime support for parallel Haskell dialects

Berthold

Loidl²,

Hammond

2016

J. Funct. Prog.

View full text Add to dashboard Cite

Over time, several competing approaches to parallel Haskell programming have emerged. Different approaches support parallelism at various different scales, ranging from small multicores to massively parallel high-performance computing systems. They also provide varying degrees of control, ranging from completely implicit approaches to ones providing full programmer control. Most current designs assume a shared memory model at the programmer, implementation and hardware levels. This is, however, becoming increasingly divorced from the reality at the hardware level. It also imposes significant unwanted runtime overheads in the form of garbage collection synchronisation etc. What is needed is an easy way to abstract over the implementation and hardware levels, while presenting a simple parallelism model to the programmer.The PAEAN (PArallEl shAred Nothing) runtime system design aims to provide a portable and high-level shared-nothing implementation platform for parallel Haskell dialects. It abstracts over major issues such as work distribution and data serialisation, consolidating existing, successful designs into a single framework. It also provides an optional virtual shared-memory programming abstraction for (possibly) shared-nothing parallel machines, such as modern multicore/manycore architectures or cluster/cloud computing systems. It builds on, unifies, and extends, existing well-developed support for shared-memory parallelism that is provided by the widely-used GHC Haskell compiler. This paper summarises the state-of-the-art in shared-nothing parallel Haskell implementations, introduces the PAEAN abstractions, shows how they can be used to implement three distinct parallel Haskell dialects, and demonstrates that good scalability can be obtained on recent parallel machines. * Corresponding author. Reported work performed while at the University of Copenhagen (DIKU).

show abstract

Section: Manticorementioning

confidence: 99%

PAEAN: Portable and scalable runtime support for parallel Haskell dialects

Berthold

Loidl²,

Hammond

2016

J. Funct. Prog.

View full text Add to dashboard Cite

show abstract

“…Frameworks like Cilk [3,9] and Wool [7,8] allow writing parallel programs in a style similar to sequential programs [1].…”

Section: Task-based Parallelismmentioning

confidence: 99%

“…Recently, Acar et al proposed two algorithms for work-stealing using private deques [1]. See further [1] for an overview of other work with private deques.…”

Section: Work-stealing Dequesmentioning

confidence: 99%

“…Acar et al write that concurrent deques suffer from two limitations: 1) local deque operations (mainly pop) require expensive memory fences in modern weakmemory architectures; 2) they can be very difficult to extend to support various optimizations, especially steal-multiple extensions [1]. They lift both limitations using private deques.…”

Section: Contributionsmentioning

confidence: 99%

“…We evaluate the performance of Lace using several benchmarks, including standard Cilk benchmarks and the UTS benchmark [13]. We compare our algorithm with Wool and with an implementation of the receiver-initiated private deque algorithm [1] in the Lace framework. Our experiments show that our algorithm is competitive with both Wool and the private deque algorithm, while lifting both limitations described in [1].…”

Section: Contributionsmentioning

confidence: 99%

See 2 more Smart Citations

Lace: Non-blocking Split Deque for Work-Stealing

Dijk

Pol

2014

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Work-stealing is an efficient method to implement load balancing in fine-grained task parallelism. Typically, concurrent deques are used for this purpose. A disadvantage of many concurrent deques is that they require expensive memory fences for local deque operations. In this paper, we propose a new non-blocking work-stealing deque based on the split task queue. Our design uses a dynamic split point between the shared and the private portions of the deque, and only requires memory fences when shrinking the shared portion. We present Lace, an implementation of work-stealing based on this deque, with an interface similar to the work-stealing library Wool, and an evaluation of Lace based on several common benchmarks. We also implement a recent approach using private deques in Lace. We show that the split deque and the private deque in Lace have similar low overhead and high scalability as Wool.

show abstract

Work stealing with private integer–vector–matrix data structure for multi‐core branch‐and‐bound algorithms

Gmys

Leroy

Mezmaz

et al. 2016

Concurrency and Computation

View full text Add to dashboard Cite

In this paper, the focus is put on multi-core branch-and-bound algorithms for solving large-scale permutationbased optimization problems. We investigate five work stealing (WS) strategies with a new data structure called integer-vector-matrix (IVM). In these strategies, each thread has a private IVM allowing the local management of a set of subproblems enumerated using a factorial system. The WS strategies differ in the way the victim thread is selected and the granularity of stolen work units (intervals of factoradics). To assess the efficiency of the private IVM-based WS approach, the five WS strategies have been extensively experimented on the flowshop scheduling permutation problem and compared with their conventional linked-list-based counterparts. The obtained results demonstrate that the IVM-based WS outperforms the linked-list-based one in terms of CPU time, memory usage and number of performed WS operations. of a process share the same virtual memory. Therefore, communications between these threads are faster than between processes. Many programming methods (such as [1-3] and [4]) based on the use of threads have been developed. To the best of our knowledge, all parallel B&B algorithms developed in the literature are based on using one or several pool(s) that store subproblems [5]. In these conventional approaches, B&B threads cooperate by adding subproblems to or removing subproblems from this or these pool(s), usually implemented as a linked-list (LL). The parallel B&B algorithm stops when this or these pool(s) is or are empty.In this paper, we propose a new approach to manage the pools of subproblems of a B&B parallel algorithm using five different work stealing (WS) strategies. Our new B&B approach aims at reducing the CPU time used to manage the thread-private B&B pools, the memory behaviour of the algorithm, the number of performed WS operations and finally the total execution time.This new approach is based on the use of an integer-vector-matrix (IVM) data structure [6] to manage the thread-private B&B pools. In our stealing strategies, work units stolen by threads are intervals of factoradics instead of sets of nodes. The factoradic, called also factorial number system, is a mixed radix numeral system adapted to numbering permutations. Based on the IVM-data structure and the factoradic-based WS, we propose a multi-core B&B algorithm and its extension to a hybrid multi-core-GPU B&B algorithm.The IVM-based B&B is compared with the LL-based B&B approach by solving some hard flowshop scheduling problem instances using the five WS strategies. The multi-core IVM-based approach spends on average 11 times less time managing the pool of subproblems, and for four out of five WS strategies, idle workers wait at least 10 times less for new work. Moreover, our experimental results indicate that, for the IVM-based approach, there is no direct relation between the number of performed WS operations and the instance size, while the number of WS events in the listed-list approach clearly grows with the size of the...

show abstract

Scheduling parallel programs by work stealing with private deques

Cited by 32 publications

References 37 publications

PAEAN: Portable and scalable runtime support for parallel Haskell dialects

PAEAN: Portable and scalable runtime support for parallel Haskell dialects

Lace: Non-blocking Split Deque for Work-Stealing

Work stealing with private integer–vector–matrix data structure for multi‐core branch‐and‐bound algorithms

Contact Info

Product

Resources

About