Cilk: An Efficient Multithreaded Runtime System

Blumofe, Robert D.; Joerg, Christopher F.; Kuszmaul, Bradley C.; Leiserson, Charles E.; Randall, Keith H.; Zhou, Yuli

doi:10.1006/jpdc.1996.0107

Cited by 922 publications

(942 citation statements)

References 32 publications

Supporting

Mentioning

924

Contrasting

Unclassified

Order By: Relevance

“…Dongarra et al proposed dynamic schedulers optimized for some linear algebra problems on general-purpose multicore processors [17]. Scheduling techniques have been proposed by several emerging programming systems such as Cilk [5], Intel Threading Building Blocks (TBB) [9], OpenMP [14], Charm++ [6] and MPI microtasking [13], etc. All these systems rely on a set of extensions to common imperative programming languages, and involve a compilation stage and runtime libraries.…”

Section: Background and Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Hierarchical Scheduling of DAG Structured Computations on Manycore Processors with Dynamic Thread Grouping

Xia

Prasanna

2010

Job Scheduling Strategies for Parallel Processing

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, scheduling such DAGs onto manycore processors remains a fundamental challenge, since synchronization across dozens of threads and preserving precedence constraints can dramatically degrade the performance. In order to improve scheduling performance on manycore processors, we propose a hierarchical scheduling method with dynamic thread grouping, which schedules DAG structured computations at three different levels. At the top level, a supermanager separates threads into groups, each consisting of a manager thread and several worker threads. The supermanager dynamically merges and partitions the groups to adapt the scheduler to the input task dependency graphs. Through group merging and partitioning, the proposed scheduler can dynamically adjust to become a centralized scheduler, a distributed scheduler or somewhere in between, depending on the input graph. At the group level, managers collaboratively schedule tasks for their workers. At the within-group level, workers perform self-scheduling within their respective groups and execute tasks. We evaluate the proposed scheduler on the Sun Ultra-SPARC T2 (Niagara 2) platform that supports up to 64 hardware threads. With respect to various input task dependency graphs, the proposed scheduler exhibits superior performance when compared with other various baseline methods, including typical centralized and distributed schedulers.

show abstract

Section: Background and Related Workmentioning

confidence: 99%

“…To compare the performance of the proposed method, we performed DAG structured computations using Charm++ [6] Cilk [5] and OpenMP [14]. In addition, we implemented three typical schedulers called Cent ded, Dist shared and Steal, respectively.…”

Section: Baselinementioning

confidence: 99%

Hierarchical Scheduling of DAG Structured Computations on Manycore Processors with Dynamic Thread Grouping

Xia

Prasanna

2010

Job Scheduling Strategies for Parallel Processing

View full text Add to dashboard Cite

show abstract

“…Of particular relevance to our work are lightweight runtimes for task parallelism such as Cilk [13], OpenMP [4] and Intel's Thread Building Blocks (TBB) [3]. These runtimes employ modern work-stealing scheduler designs similar to our own, but do not provide primitives suitable for implementing process-oriented designs.…”

Section: Related Workmentioning

confidence: 99%

Multicore Scheduling for Lightweight Communicating Processes

Ritson

Sampson

Barnes

2009

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Abstract. Process-oriented programming is a design methodology in which software applications are constructed from communicating concurrent processes. A process-oriented design is typically composed of a large number of small isolated concurrent components. These components allow for the scalable parallel execution of the resulting application on both shared-memory and distributed-memory architectures. In this paper we present a runtime designed to support process-oriented programming by providing lightweight processes and communication primitives. Our runtime scheduler, implemented using lock-free algorithms, automatically executes concurrent components in parallel on multicore systems. Runtime heuristics dynamically group processes into cache-affine work units based on communication patterns. Work units are then distributed via wait-free work-stealing. Initial performance analysis shows that, using the algorithms presented in this paper, process-oriented software can execute with an efficiency approaching that of optimised sequential and coarse-grain threaded designs.

show abstract

“…We use initial rough estimation of the combined computational power of each cluster involved in the computation (based on CPU clock speed) for the weighted partitioning of the work-pool and initial assignment of work. However, this is a rough "guess" estimation, which is adjusted at runtime using a combination of master/worker and work-stealing [6,7] methods. Each master has a copy of the global work-pool, which are identical in the beginning of the computation.…”

Section: Multi-level Dynamic Loadmentioning

confidence: 99%

Grid-Enabled Software Environment for Enhanced Dynamic Data-Driven Visualization and Navigation During Image-Guided Neurosurgery

Chrisochoides

Fedorov

Kot

et al. 2007

Computational Science – ICCS 2007

View full text Add to dashboard Cite

Abstract. In this paper we present our experience with an Image Guided Neurosurgery Grid-enabled Software Environment (IGNS-GSE) which integrates real-time acquisition of intraoperative Magnetic Resonance Imaging (IMRI) with the preoperative MRI, fMRI, and DT-MRI data. We describe our distributed implementation of a non-rigid image registration method which can be executed over the Grid. Previously, non-rigid registration algorithms which use landmark tracking across the entire brain volume were considered not practical because of the high computational demands. The IGNS-GSE, for the first time ever in clinical practice, alleviated this restriction. We show that we can compute and present enhanced MR images to neurosurgeons during the tumor resection within minutes after IMRI acquisition. For the last 12 months this software system is used routinely (on average once a month) for clinical studies at Brigham and Women's Hospital in Boston, MA. Based on the analysis of the registration results, we also present future directions which will take advantage of the vast resources of the Grid to improve the accuracy of the method in places of the brain where precision is critical for the neurosurgeons.

show abstract

Cilk: An Efficient Multithreaded Runtime System

Cited by 922 publications

References 32 publications

Hierarchical Scheduling of DAG Structured Computations on Manycore Processors with Dynamic Thread Grouping

Hierarchical Scheduling of DAG Structured Computations on Manycore Processors with Dynamic Thread Grouping

Multicore Scheduling for Lightweight Communicating Processes

Grid-Enabled Software Environment for Enhanced Dynamic Data-Driven Visualization and Navigation During Image-Guided Neurosurgery

Contact Info

Product

Resources

About