Title :
An adaptive global reduction algorithm for wormhole-routed 2D meshes
Author :
Huang, Yih ; McKinley, Philip K.
Author_Institution :
Dept. of Comput. Sci., Michigan State Univ., East Lansing, MI, USA
Abstract :
This paper presents a global reduction algorithm for wormhole-routed 2D meshes. Well-known reduction algorithms that are optimized for short vectors have complexity O(M log N), where N=n×n is the number of nodes, and M the vector length. Algorithms suitable for long vectors have complexity O(√N+M). Previously known asymptotically optimal algorithms with complexity O(log N+M) incur inherent network contention among constituent messages. The proposed algorithm adapts to the given vector length, resulting in complexities O(M log N) for short vectors, O(log N+M) for medium-sized vectors, and O(√N+M) for sufficiently long vectors. The O(√N+M) version is preferred to the O(log N+M) version for long vectors, due to its small coefficient associated with M, the dominating factor for such vectors. The algorithm is contention-free in a synchronous environment. Under asynchronous execution models, depth contention (contention among message-passing steps) may occur. Our simulation studies show that the effect of depth contention on the actual performance is negligible
Keywords :
computational complexity; hypercube networks; message passing; adaptive global reduction algorithm; asymptotically optimal algorithms; asynchronous execution models; complexity; depth contention; dominating factor; inherent network contention; message-passing; reduction algorithms; short vectors; simulation studies; wormhole-routed 2D meshes; Computational modeling; Computer networks; Computer science; Computer simulation; Concurrent computing; Network topology; Performance loss; Pipeline processing; Telecommunication traffic; US Department of Energy;
Conference_Titel :
Parallel and Distributed Processing, 1995. Proceedings. Seventh IEEE Symposium on
Conference_Location :
San Antonio, TX
Print_ISBN :
0-81867195-5
DOI :
10.1109/SPDP.1995.530673