Hybrid OpenCL: Enhancing OpenCL for Distributed Processing

Aoki, Ryo; Oikawa, Shuichi; Nakamura, Takashi; Miki, Satoshi

doi:10.1109/ispa.2011.28

Cited by 25 publications

(19 citation statements)

References 4 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Our MC hybrid design approach uses OpenCL code to use parallel implementation in which part of the computation task is off-loaded to CPU while GPU is running other computation task in parallel. Aoki et al (2011) presented Hybrid OpenCL implementation for multiple nodes in network environment. The concept is similar to what we presented in this study between GPU and CPU, but in (Aoki et al, 2011) the performance is compared between Hybrid Open CL and OpenCL with MPI implementation.…”

Section: Related Workmentioning

confidence: 99%

Hybrid Implementation and Performance Analysis for High Performance Computation Workload

Issa¹

2014

Journal of Computer Science

View full text Add to dashboard Cite

Given the need to achieve maximum performance possible, offloading intensive computation workload to GPU is a key to achieve this goal. Offloading most of the workload to GPU may not results in desired performance, so a middle approach is more suitable such as splitting the workload between the CPU and the GPU can be considered as an optimized approach. In this study, we used a popular high performance computation workload which can also be implemented using a hybrid approach in which part of the workload is offloaded to the CPU. We also present a performance estimation method which is verified to estimate performance with in 5% error margin.

show abstract

Section: Related Workmentioning

confidence: 99%

Hybrid Implementation and Performance Analysis for High Performance Computation Workload

Issa¹

2014

Journal of Computer Science

View full text Add to dashboard Cite

show abstract

“…When the program is executed, the runtime features are provided to the previously trained model (5), which combines them with the static program features to predict the best task partitioning for the current program with the selected problem size (6). Finally, the runtime system executes the program on the given hardware using the predicted task partitioning (7).…”

Section: Architecturementioning

confidence: 99%

“…Several projects [32,6,9,31,24,23] mainly focused on OpenMP, CUDA, and OpenCL extensions, investigated how to facilitate the programming of clusters with heterogeneous nodes. Our work, while following the same idea, targets an automatic management of multiple devices in a single node.…”

Section: Related Workmentioning

confidence: 99%

An automatic input-sensitive approach for heterogeneous task partitioning

Kofler

Grasso

Cosenza

et al. 2013

Proceedings of the 27th International ACM Conference on International Conference on Supercomputing

View full text Add to dashboard Cite

Kofler, K.; Grasso, I.; Cosenza, B.; Fahringer, T.: An automatic input-sensitive approach for heterogeneous task partitioning. ABSTRACTUnleashing the full potential of heterogeneous systems, consisting of multi-core CPUs and GPUs, is a challenging task due to the difference in processing capabilities, memory availability, and communication latencies of different computational resources.In this paper we propose a novel approach that automatically optimizes task partitioning for different (input) problem sizes and different heterogeneous multi-core architectures. We use the Insieme source-to-source compiler to translate a single-device OpenCL program into a multi-device OpenCL program. The Insieme Runtime System then performs dynamic task partitioning based on an offline-generated prediction model. In order to derive the prediction model, we use a machine learning approach based on Artificial Neural Networks (ANN) that incorporates static program features as well as dynamic, input sensitive features. Principal component analysis have been used to further improve the task partitioning. Our approach has been evaluated over a suite of 23 programs and respectively achieves a performance improvement of 22% and 25% compared to an execution of the benchmarks on a single CPU and a single GPU which is equal to 87.5% of the optimal performance.

show abstract

“…rCUDA [11] provides this functionality for Nvidia devices and Mosix VCL [12] provides this functionality for OpenCL on Linux-based systems. Hybrid OpenCL [13], dOpenCL [14], and Distributed OpenCL [15] provide this functionality cross vendor and cross platform. In clOpenCL [16] each remote node is represented locally as a separate platform.…”

Section: E Snuclmentioning

confidence: 99%

DistCL: A Framework for the Distributed Execution of OpenCL Kernels

Diop

Gurfinkel

Anderson

et al. 2013

2013 IEEE 21st International Symposium on Modelling, Analysis and Simulation of Computer and Telecommunication Systems

View full text Add to dashboard Cite

Abstract-GPUs are used to speed up many scientific computations; however, to use several networked GPUs concurrently, the programmer must explicitly partition work and transmit data between devices. We propose DistCL, a novel framework that distributes the execution of OpenCL kernels across a GPU cluster. DistCL makes multiple distributed compute devices appear to be a single compute device. DistCL abstracts and manages many of the challenges associated with distributing a kernel across multiple devices including: (1) partitioning work into smaller parts, (2) scheduling these parts across the network, (3) partitioning memory so that each part of memory is written to by at most one device, and (4) tracking and transferring these parts of memory. Converting an OpenCL application to DistCL is straightforward and requires little programmer effort. This makes it a powerful and valuable tool for exploring the distributed execution of OpenCL kernels. We compare DistCL to SnuCL, which also facilitates the distribution of OpenCL kernels. We also give some insights: distributed tasks favor more compute bound problems and favour large contiguous memory accesses. DistCL achieves a maximum speedup of 29.1 and average speedups of 7.3 when distributing kernels among 32 peers over an Infiniband cluster.

show abstract

Hybrid OpenCL: Enhancing OpenCL for Distributed Processing

Cited by 25 publications

References 4 publications

Hybrid Implementation and Performance Analysis for High Performance Computation Workload

Hybrid Implementation and Performance Analysis for High Performance Computation Workload

An automatic input-sensitive approach for heterogeneous task partitioning

DistCL: A Framework for the Distributed Execution of OpenCL Kernels

Contact Info

Product

Resources

About