Active messages: a mechanism for integrated communication and computation
ISCA '92 Proceedings of the 19th annual international symposium on Computer architecture
Graph theory and its applications
Graph theory and its applications
Genetic Algorithms in Search, Optimization and Machine Learning
Genetic Algorithms in Search, Optimization and Machine Learning
Introduction to Algorithms
LLM: A Low Latency Messaging Infrastructure for Linux Clusters
HiPC '02 Proceedings of the 9th International Conference on High Performance Computing
Exploting communication Latency Hiding for Parallel Network
Proceedings of the 1994 International Conference on Parallel and Distributed Systems
Identifying the Capability of Overlapping Computation with Communication
PACT '96 Proceedings of the 1996 Conference on Parallel Architectures and Compilation Techniques
Strategies for Achieving High Performance Incremental Computing on a Network Environment
AINA '04 Proceedings of the 18th International Conference on Advanced Information Networking and Applications - Volume 2
Bounds on the Client-Server Incremental Computing
IEICE Transactions on Fundamentals of Electronics, Communications and Computer Sciences
| Hi-index | 0.00 |
Network latency is an adverse factor for computations performed across the network. Overlapping computation with communication is an important technique for hiding latency. It has been shown that network latency cannot be effectively hidden without considering the order of sending data [C.-C. Lin, Strategies for achieving high performance incremental computing on a network environment, in: Proc.18th Int'l Conf. on Advanced Information Networking and Applications 1, 2004, pp. 113-118]. However, finding a data-sending order for the input to a task which minimizes the remote execution time for any network traffic pattern is NP-hard [C.-C. Lin, D.-W. Wang, T.-S. Hsu, Bounds on the client-server incremental computing, IEICE Trans. Fundamentals E89-A (5) (2006) 1198-1206]. Thus, heuristic algorithms are often employed to search an optimal input stream. The performance of algorithms relies on an effective mechanism for guiding the search toward a promising direction. In this paper, the computation-progress graph is proposed for transforming an input stream of a task to its corresponding pattern of progressive computations. Then, the assessing function is defined for assigning scores to the found input streams based on the computation-progress graph. Based on the scores, the promising search directions can be determined. Finally, the effectiveness of our assessing function is also demonstrated by the search of the optimal orders for computing the product of two polynomials, matrix multiplication and FFT.