An Improved Performance by Low Rerouting Hops and Fault-Tolerant Guaranteeing Dynamic Rerouting Multistage Interconnection Network

C.-W. Chen and S.-C. Fu (Taiwan)

Keywords

Parallel computing, fault-tolerant, Gammainterconnection network (GIN), dynamic rerouting,collision, and throughput

Abstract

Disjoint paths and dynamic rerouting are often used in the design of fault-tolerant interconnection networks to provide fault tolerance in parallel computing systems. Disjoint paths can guarantee fault-tolerance, but rerouting packets and sending more packets at the same time require additional hardware and cause a drop in performance. The dynamic rerouting method provides alternative paths to a destination when a packet meets a faulty element. In this paper, we propose a dynamic rerouting gamma interconnection network (DRGIN) that provides dynamic rerouting capability to guarantee one fault tolerance. DRGIN uses the dynamic rerouting method to reroute packets when a collision occurs. To achieve high performance, DRGIN offers a fixed number of rerouting hops (2) and the same hardware cost as a Gamma network. More importantly, DRGIN ensures that a packet after rerouting will not encounter the same faulty element again requiring yet another rerouting. For these reasons, and from our experimental results, DRGIN provides a better throughput than other fault-tolerant networks..

Important Links:



Go Back