Exploiting coarse-grained task, data, and pipeline parallelism in stream programs

Gordon, Michael I.; Thies, William; Amarasinghe, Saman

doi:10.1145/1168919.1168877

Cited by 100 publications

(159 citation statements)

References 31 publications

Supporting

Mentioning

159

Contrasting

Order By: Relevance

“…However, data parallelism has the hazard of increasing buffering and latency, and the limitation of being unable to parallelize nodes in an active state [18]. In other words, an appropriate granularity of data parallelism is needed to match the data characteristics.…”

Section: Adjusting the Granularity Of Data Parallelismmentioning

confidence: 99%

“…The key characteristics of the benchmarks are depicted in Table 5. The automic data parallelism and coased-grained parallelism [18] technology were choosed in the comparison as dynamic approaches. In order to ensure the consistency of test platform, we implemented the benchmarks on the messagedriven based stream programming framework [50].…”

Section: Adapting To Data Stream Fluctuationsmentioning

confidence: 99%

See 1 more Smart Citation

Achieving self-aware parallelism in stream programs

Feng

Talpur

et al. 2014

Cluster Comput

View full text Add to dashboard Cite

The age of big data open the door to a new approach in data exploration and utilization. With the increasing complexities and dynamics of modern IT systems and services, it has become a challenge to effectively exploit parallelism on multicore platforms in computing systems that are heterogeneous, dynamic and decentralised. Self-aware software is a response to these demands in dealing with distributed applications in changing environments. It is a closedloop system with a series of optimization strategies to adjust itself dynamicly during data processing. We focus on incorporating adaptation mechanisms into the stream programs for exposing distributed parallelism. In the traditional stream programming models, changing data and status normally require human supervision to adjust the stream graph for performance. As one-time optimization strategy, the reconfiguration and maintenance lead to costly and time-consuming procedures during the operating phase. To address these problems, we propose a self-aware stream programming model called StreamAware. A key property of this model is that exposing self-aware parallelism in the message driven execution paradigm, which provides dynamic and reconfigurable stream graph in adapting to the data flow changes. The model defines the self-awareness loop based on finite state machine for stream applications to adjust their own stream graph with continuous optimization strategy. This paper presents three different self-aware systems built using StreamAware. The empirical evaluation demonstrate how these systems can exploit self-aware parallelism using the Parsec benchmark problems, optimize performance per Watt, and respond to significant changes in stream processing.

show abstract

Section: Adjusting the Granularity Of Data Parallelismmentioning

confidence: 99%

Section: Adapting To Data Stream Fluctuationsmentioning

confidence: 99%

Achieving self-aware parallelism in stream programs

Feng

Talpur

et al. 2014

Cluster Comput

View full text Add to dashboard Cite

show abstract

“…Although it is similar to our idea of decoupling/pipelining computation with memory, the underline model in inspector-executor is taskdriven, whereas the computation tasks determine the schedule of data communication/prefetching. -Streaming programming: In [13,20] the authors performed a comprehensive study of regular/irregular scientific computing applications on streaming programming model. Both their work and ours share the streaming programming style of gathercompute-scatter.…”

Section: Related Workmentioning

confidence: 99%

Analysis and performance results of computing betweenness centrality on IBM Cyclops64

Tan

Sreedhar²,

Gao

2009

J Supercomput

View full text Add to dashboard Cite

This paper presents a joint study of application and architecture to improve the performance and scalability of an irregular application-computing betweenness centrality-on a many-core architecture IBM Cyclops64. The characteristics of unstructured parallelism, dynamically non-contiguous memory access, and low arithmetic intensity in betweenness centrality pose an obstacle to an efficient mapping of parallel algorithms on such many-core architectures. By identifying several key architectural features, we propose and evaluate efficient strategies for achieving scalability on a massive multi-threading many-core architecture. We demonstrate several optimization strategies including multi-grain parallelism, just-in-time locality with explicit memory hierarchy and non-preemptive thread execution, and fine-grain data synchronization. Comparing with a conventional parallel algorithm, we get 4X-50X improvement in performance and 16X improvement in scalability on a 128-cores IBM Cyclops64 simulator.

show abstract

“…in the domains regarding signal processing, multimedia and graphics [5][6][7]. Yet it has not been sufficiently validated whether stream processor is efficient for scientific computing.…”

Section: Fig 1 Ft64 Stream Architecturementioning

confidence: 99%

Matrix-based streamization approach for improving locality and parallelism on FT64 stream processor

Yang

Yan

et al. 2008

J Supercomput

View full text Add to dashboard Cite

FT64 is the first 64-bit stream processor designed for scientific computing. It is critical to exploit optimizing streamization approaches for scientific applications on FT64 due to the inefficiency of direct streamization approach. In this paper, we propose a novel matrix-based streamization approach for improving locality and parallelism of scientific applications on FT64. First, a Data&Computation Matrix is built to abstract the relationship between loops and arrays of the original programs, and it is helpful for formulating the streamization problem. Second, three key techniques for optimizing streamization approach are proposed based on the transformations of the matrix, i.e., coarse-grained program transformations, fine-grained program transformations, and stream organization optimizations. Finally, we apply our approach to ten typical scientific application kernels on FT64. The experimental results show that the matrix-based streamization approach achieves an average speedup of 2.76 over the direct streamization approach, and performs equally to or better than the corresponding Fortran programs on Itanium 2 except CG. It is certain that the matrix-based streamization approach is a promising and practical solution to efficiently exploit the tremendous potential of FT64.

show abstract

Exploiting coarse-grained task, data, and pipeline parallelism in stream programs

Cited by 100 publications

References 31 publications

Achieving self-aware parallelism in stream programs

Achieving self-aware parallelism in stream programs

Analysis and performance results of computing betweenness centrality on IBM Cyclops64

Matrix-based streamization approach for improving locality and parallelism on FT64 stream processor

Contact Info

Product

Resources

About