Executing task graphs using work-stealing

Agrawal, Kunal; Leiserson, Charles E.; Sukha, Jim

doi:10.1109/ipdps.2010.5470403

Cited by 55 publications

(50 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The Nabbit [11] library extension to Cilk++ allows the user to declare arbitrary task dependences so as to create task graphs that cannot be created with nested fork-join operations. Unlike DDTs, where the task graph is specified implicitly via await clauses on async statements, it is created explicitly in Nabbit via calls to an AddDep() method before the start of graph execution.…”

Section: Resultsmentioning

confidence: 99%

“…The sets of task graphs that can be supported by a dynamic scheduler depend on the underlying task primitives in the parallel programming model e.g., the task graphs created by Cilk's spawn and sync primitives are restricted to fully-strict computation graphs [8], whereas the task graphs created by async and finish constructs in X10 and HJ belong to the set of terminally-strict computation graphs [9]. A number of researchers have advocated the benefits of more general task graph structures [10], [11], [12], and shown that the use of these task graph structures can lead to improved performance.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Data-Driven Tasks and Their Implementation

Taşırlar

Sarkar

2011

2011 International Conference on Parallel Processing

View full text Add to dashboard Cite

Dynamic task parallelism has been identified as a prerequisite for improving productivity and performance on future many-core processors. In dynamic task parallelism, computations are created dynamically and the runtime scheduler is responsible for scheduling the computations across processor cores. The sets of task graphs that can be supported by a dynamic scheduler depend on the underlying task primitives in the parallel programming model, with various classes of fork-join structures used most often in practice. However, many researchers have advocated the benefits of more general task graph structures, and have shown that the use of these task graph structures can lead to improved performance.In this paper, we propose an extension to task parallelism called Data-Driven Tasks (DDTs) that can be used to create arbitrary task graph structures. Unlike a normal task that starts execution upon creation, a DDT specifies its input constraints in an await clause containing a list of Data-Driven Futures (DDFs). A DDF can be viewed as a container with a full/empty state that obeys a dynamic single-assignment rule. The runtime scheduler will then ensure that a task is only scheduled when all the DDFs in its await clause become available (full). There is no constraint on which task performs a put() operation on a DDF.We describe five scheduling algorithms (Coarse-Grain Blocking, Fine-Grain Blocking, Delayed Async, Rollback & Replay, Data-Driven) that can be used to implement DDTs, and include performance evaluations of these five algorithms on a variety of benchmark programs and multi-core platforms. Our results show that the Data-Driven scheduler is the best approach for implementing DDTs, both from the viewpoints of memory efficiency and scalable parallelism.

show abstract

Section: Resultsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Data-Driven Tasks and Their Implementation

Taşırlar

Sarkar

2011

2011 International Conference on Parallel Processing

View full text Add to dashboard Cite

show abstract

“…Work-stealing (e.g., [34]) is a concept related to adaptive load balancing, but with important differences. Rather than moving the partitions among the machines, it implies dividing the workload into many more partitions than the number of available machines.…”

Section: Monotonic Joinsmentioning

confidence: 99%

Load balancing and skew resilience for parallel joins

Vitorovic

Elseidy

2016

2016 IEEE 32nd International Conference on Data Engineering (ICDE)

View full text Add to dashboard Cite

Abstract-We address the problem of load balancing for parallel joins. We show that the distribution of input data received and the output data produced by worker machines are both important for performance. As a result, previous work, which optimizes either for input or output, stands ineffective for load balancing. To that end, we propose a multi-stage load-balancing algorithm which considers the properties of both input and output data through sampling of the original join matrix. To do this efficiently, we propose a novel category of equi-weight histograms. To build them, we exploit state-of-the-art computational geometry algorithms for rectangle tiling. To our knowledge, we are the first to employ tiling algorithms for join load-balancing. In addition, we propose a novel, join-specialized tiling algorithm that has drastically lower time and space complexity than existing algorithms. Experiments show that our scheme outperforms stateof-the-art techniques by up to a factor of 15.

show abstract

“…To the underlying fork-join pattern, each node meets the single predecessor requirement by having its last-completing task graph predecessor function as its parent in the fork-join pattern. Note that Nabbit [12] uses Cilk++ to execute arbitrary task graphs in much the same way.…”

Section: B Fork-join Parallel Patternmentioning

confidence: 99%

Executing Optimized Irregular Applications Using Task Graphs within Existing Parallel Models

Krieger

Strout

Roelofs

et al. 2012

2012 SC Companion: High Performance Computing, Networking Storage and Analysis

View full text Add to dashboard Cite

Abstract-Many sparse or irregular scientific computations are memory bound and benefit from locality improving optimizations such as blocking or tiling. These optimizations result in asynchronous parallelism that can be represented by arbitrary task graphs. Unfortunately, most popular parallel programming models with the exception of Threading Building Blocks (TBB) do not directly execute arbitrary task graphs. In this paper, we compare the programming and execution of arbitrary task graphs qualitatively and quantitatively in TBB, the OpenMP doall model, the OpenMP 3.0 task model, and Cilk Plus. We present performance and scalability results for 8 and 40 core shared memory systems on a sparse matrix iterative solver and a molecular dynamics benchmark.

show abstract

Executing task graphs using work-stealing

Cited by 55 publications

References 17 publications

Data-Driven Tasks and Their Implementation

Data-Driven Tasks and Their Implementation

Load balancing and skew resilience for parallel joins

Executing Optimized Irregular Applications Using Task Graphs within Existing Parallel Models

Contact Info

Product

Resources

About