Parallel Prefix Sum Algorithm on Optoelectronic Biswapped Network Hyper Hexa-cell

Abstract

The biswapped network hyper hexa-cell is recently reported optoelectronic network architecture for delivering excellent performance especially for mapping numerical problems which demands frequent routing and broadcasting. This network contains some important benefits such as smaller diameter, higher bisection width, and lower network's total and optical cost as compared to counter-part OTIS hyper hexa-cell network. It is also advantageous as compared to the traditional biswapped network mesh containing smaller diameter and higher minimum node degree. In this paper, we present a parallel algorithm for mapping prefix sum of × ( × − ) data elements on a d h -dimensional biswapped network hyper hexa-cell of × ( × − ) processors (assuming each processor contain single data element). It demands total (( − ) × ( + )) − ( − )×( − ) + + 9 intra-cluster (electronic) and 3 inter-cluster (optical) moves. Index Terms-Optoelectronic, OTIS mesh, OTIS hyper hexa-cell, Biswapped network hyper hexa-cell.

Load Balancing Problem on Hyper Hexa Cell Interconnection Network

“…Several experiments have been conducted to compute the time required to execute the proposed load balancing method, on 6,12,24,48,96,192,384,768,1536 and 3072-processor HHC. That indicates that the experiments were performed for several dimensions starting from dimension one up to dimension ten.…”

“…One of these interconnection networks is Hyper Hexa Cell, which was built on the advantageous features of the hypercube interconnection network and had attractive topological properties such as diameter, minimum node degree, width of bisection and optical cost [3]. This encouraged the researchers to use this new architecture in solving many parallel algorithms in different fields, such as broadcast communication [4], unicast routing [5] and parallel prefix sum [6]. Moreover the optical links version of this interconnection network namely OTIS-HHC, was exploited by the researchers in order to solve various problems, such as communication algorithms [7], shortest path routing on OTIS-HHC [8], parallel heuristic local search algorithms [9], routing and sorting algorithms [10], parallel quick sort algorithm [11], and traveling salesman problem [12].…”