W. Liu, C. Wang, and K. Prasanna, "Portable and Scalable Algorithms for Irregular All--to--all Communication," 16th ICDCS, pages 428-435, 1996. Home/Search Document Details and Download
Summary Related Articles Check |
This paper is cited in the following contexts:
How Helpers Hasten h-Relations - Sanders, Solis-Oba (2000) (2 citations) (Correct)
.... the problem (when a PE knows only the messages that it must send, but it does not know the lengths of the messages that other PE s must send) there are several algorithms that transform an arbitrary instance of the h relation problem into two almost regular h relation problems, with h h [16, 23]. For all of these algorithms the communication time is about 2h. If h max is the maximum number of packets that some PE sends or receives, and if we assume the full duplex model of communication where each PE can simultaneously send and receive one packet, then the problem can be solved optimally ....
W. Liu, C. Wang, and K. Prasanna. Portable and scalable algorithms for irregular all-to-all communication. In 16th ICDCS, pages 428--435. IEEE, 1996.
How Helpers Hasten h-Relations - Sanders, Solis-Oba (2000) (2 citations) (Correct)
....every message m ij has length h= P Gamma 1) We show that this problem can be solved using a 1 factorization [6] of the transfer graph. There are several algorithms that transform an arbitrary instance of the h relation problem into two almost regular h relation problems, with h h [19, 14]. In the resulting schedule, almost every packet is forwarded so that the communication time is about 2h. If h is the maximum number of packets that some PE sends or receives, and if we assume the duplex model of communication where each PE can send and receive one packet simultaneously, then ....
....of a collection of disjoint cycles and paths it is easy to solve it using communication time three. Therefore, the original h relation can be solved using communication time at most 3 dh=2e. This is an improvements over algorithms which transform a general h relation into two regular h relations [19, 14], and hence need communication time at least 2h. Furthermore, since at least h rounds are needed to send the messages, this algorithm achieves a performance ratio no worse than 3dh=2e ) The translation algorithm first converts an instance of the problem into an edge coloring problem on a ....
[Article contains additional citation context not shown here]
W. Liu, C. Wang, and K. Prasanna. Portable and scalable algorithms for irregular all-to-all communication. In 16th ICDCS, pages 428--435. IEEE, 1996.
How Helpers Hasten h-Relations - Sanders, Solis-Oba (2000) (2 citations) (Correct)
.... problem (when a PE knows only the messages that it must send, but it does not know the lengths of the messages that other PE s must send) there are several algorithms that transform an arbitrary instance of the h relation problem into two almost regular h 0 relation problems, with h h 0 [16, 23]. For all of these algorithms the communication time is about 2h. If h max is the maximum number of packets that some PE sends or receives, and if we assume the full duplex model of communication where each PE can simultaneously send and receive one packet, then the problem can be solved ....
W. Liu, C. Wang, and K. Prasanna. Portable and scalable algorithms for irregular all-to-all communication. In 16th ICDCS, pages 428--435. IEEE, 1996.
Design of Application Software for Embedded Signal Processing - Liu, Prasanna Self-citation (Liu Prasanna) (Correct)
....A computational model is an analytical representation of a computer system. The performance of an algorithm architecture pair can be estimated using such a computational model. Recently, the Generalpurpose Distributed Memory (GDM) model has been used in developing ESP applications on HPC platforms [14, 18]. Explicit message passing is assumed in the GDM model for communication among the processors. This model can be used to choose appropriate algorithms to minimize the communication time when an ESP application is parallelized. Current HPC platforms employ high bandwidth low latency ....
....time when an ESP application is parallelized. Current HPC platforms employ high bandwidth low latency interconnection networks. In these platforms, large software overheads are incurred as a message traverses through various software layers and the network interface to its destination [18]. The software overheads dominate the hardware switch latencies in current HPC platforms. For algorithm design and analysis, the networks of such platforms 13 ESP Example( MPI Comm rank(MPI COMM WORLD, my id ) if(my id= P0) Execute tasks mapped onto P0 . FFT0 ....
W. Liu, C. Wang, and V. K. Prasanna, "Portable and Scalable Algorithms for Irregular All-to-all Communication," in proceedings of the 16th International Conference on Distributed Computing Systems (ICDCS '96), Hong-Kong, May 1996.
Parallel Implementation of a Class of Adaptive Signal.. - Lee, Liu, Prasanna Self-citation (Liu Prasanna) (Correct)
No context found.
W. Liu, C. Wang, and V. K. Prasanna, "Portable and Scalable Algorithms for Irregular All-to-all Communication," in proceedings of the 16th International Conference on Distributed Computing Systems (ICDCS '96), May 1996. 49
Message Scheduling for All-to-all Personalized Communication on .. - Faraj, Yuan (Correct)
No context found.
W. Liu, C. Wang, and K. Prasanna, "Portable and Scalable Algorithms for Irregular All--to--all Communication," 16th ICDCS, pages 428-435, 1996.
Online articles have much greater impact More about CiteSeer.IST Add search form to your site Submit documents Feedback
CiteSeer.IST - Copyright Penn State and NEC