A Hierarchical Load-Balancing Framework for Dynamic Multithreaded Computations

Recent trend has made it clear that the processor makers are committed to the multicore chip designs. The number of cores per chip is increasing, while there is little or no increase in the clock speed. This parallelism trend poses a significant and urgent challenge on computer software because programs have to be written or transformed into a multi-threaded form to take full advantage of future hardware advances.Task parallelism has been identified as one of the prerequisites for software productivity. In task parallelism, programmers focus on decomposing the problem into subcomputations that can run in parallel and leave the compiler and runtime to handle the scheduling details. This separation of concerns between task decomposition and scheduling provides productivity to the programmer but poses challenges to the runtime scheduler.Our thesis is that work-stealing schedulers with adaptive scheduling policies and locality-awareness can provide a scalable and robust runtime foundation for multicore task parallelism. We evaluate our thesis using the new Scalable Locality-aware Adaptive Work-stealing (SLAW) runtime scheduler developed for the Habanero-Java programming language, a task-parallel variant of Java. SLAW's adaptive task scheduling is motivated by the study of two common scheduling policies in a work-stealing scheduler, specifically, the work-first and the help-first policy. Both policies exhibit limitations in performance and resource usage in different situations. The variances make it hard to determine the best policy a priori. SLAW addresses these limitations by supporting both policies simultaneously and selecting policies adaptively on a per-task basis at runtime. Our results show that SLAW achieves 0.98× to 9.2× speedup over the help-first scheduler and 0.97× to 4.5× speedup over the work-first scheduler. Further, for large irregular parallel computations, SLAW supports data sizes and achieves performance that cannot be delivered by the use of any single fixed policy.SLAW's locality-aware scheduling framework aims to overcome the cache unfriendliness of work-stealing due to randomized stealing. The SLAW scheduler is designed for programming models where locality hints are provided to the runtime by the programmer or compiler. Our results show that locality-aware scheduling can improve performance by increasing temporal data reuse for iterative data-parallel applications.iii

show abstract

“…The locality issue in multithreaded computation has also received a lot of attention in past work [3,19,51,61,91,72,13,12].…”

Section: Research In Task Parallelismmentioning

confidence: 99%

SLAW: A scalable locality-aware adaptive work-stealing scheduler

Zhao

Cavé

et al. 2010

2010 IEEE International Symposium on Parallel &Amp; Distributed Processing (IPDPS)

View full text Add to dashboard Cite

show abstract

“…However, the RTS can easily be extended to support hierarchical work stealing [12] or algorithms based on peer-to-peer techniques [8], for example. When starting an application, the desired work stealing algorithm can be selected by providing a parameter to the Constellation RTS.…”

Section: Constellation Run Time Systemmentioning

confidence: 99%

Towards jungle computing with Ibis/Constellation

Maassen¹,

Drost²,

Bal³

et al. 2011

Proceedings of the 2011 Workshop on Dynamic Distributed Data-Intensive Applications, Programming Abstractions, and Systems

View full text Add to dashboard Cite

The scientific computing landscape is becoming more and more complex. Besides traditional supercomputers and clusters, scientists can also apply grid and cloud infrastructures. Moreover, the current integration of many-core technologies such as GPUs with such infrastructures adds to the complexity. To make matters worse, data distribution, hardware availability, software heterogeneity, and increasing data sizes, commonly force scientists to use multiple computing platforms simultaneously: a true computing jungle.In this paper we introduce Ibis/Constellation, a software platform specifically designed for distributed, heterogeneous and hierarchical computing environments. In Ibis/Constellation we assume that applications consist of several distinct (but somehow related) activities. These activities can be implemented independently using existing, well understood tools (e.g. MPI, CUDA, etc.). Ibis/Constellation is then used to construct the overall application by coupling the distinct activities. Using application defined labels in combination with context-aware work stealing, Ibis/Constellation provides a simple and efficient mechanism for automatically mapping the activities to the appropriate resources, taking data locality and heterogeneity into account.We show that an existing supernova detection application can be ported to Ibis/Constellation with little effort. By making small changes to the application defined labels, this example application can run efficiently in three very different HPC computing environments: a distributed set of clusters, a large 48-core machine, and a GPU cluster.

show abstract

“…In general, however, programs come in a variety of structures [2]. Karamcheti and Chien studied hierarchical load balancing framework for multithreaded computations for employing various scheduling policies for a system [16]. Recent research on scheduling DAGs includes [17] where the authors studied the problem of scheduling more than one DAG simultaneously onto a set of heterogeneous resources, and [1] where Ahmad proposed a game theory based scheduler on multicore processors for minimizing energy consumption.…”

Section: Background and Related Workmentioning

confidence: 99%

Exploring Weak Dependencies in DAG Scheduling

Xia

Prasanna

2011

2011 IEEE International Symposium on Parallel and Distributed Processing Workshops and PHD Forum

View full text Add to dashboard Cite

Abstract-Many computational solutions can be expressed as directed acyclic graphs (DAGs) with weighted nodes. In parallel computing, a fundamental challenge is to efficiently map computing resources to the tasks, while preserving the precedence constraints among the tasks. Traditionally, such constraints are preserved by starting a task after all its preceding tasks are completed. However, for a class of DAG structured computations, a task can be partially executed with respect to each preceding task. We define such relationship between the tasks as weak dependency. In this paper, we adapt a traditional DAG scheduling scheme to exploit weak dependencies in a DAG. We perform experiments to study the impact of weak dependency based scheduling method on the execution time using a representative set of task graphs for exact inference in junction trees. For a class of task graphs, on a state-of-the-art general-purpose multicore system, the weak dependency based scheduler runs 4x faster than a baseline scheduler that is based on the traditional scheduling method.

show abstract

A Hierarchical Load-Balancing Framework for Dynamic Multithreaded Computations

Cited by 11 publications

References 23 publications

SLAW: A scalable locality-aware adaptive work-stealing scheduler

SLAW: A scalable locality-aware adaptive work-stealing scheduler

Towards jungle computing with Ibis/Constellation

Exploring Weak Dependencies in DAG Scheduling

Contact Info

Product

Resources

About