A Low Cost Fault Tolerant Packet Routing for Parallel Computers

This work presents a new switching mechanism to tolerate arbitrary faults in interconnection networks with a negligible implementation cost. Although our routing technique can be applied to any regular or irregular topology, in this paper we focus on its application to k-ary n-cube networks when managing both synthetic and real traffic workloads. Our mechanism is effective regardless the number of faults and their configuration. When the network is working without any fault, no overhead is added to the original routing scheme. In the presence of a low number of faults, the network sustains a performance close to that observed under fault-free conditions. Finally, when the number of faults increases, the system exhibits a graceful performance degradation.

• Type: Conference
• Year: 2003
• Where: IPPS
• Authors: Valentin Puente, José A. Gregorio, Ramón Beivide, Fernando Vallejo
• Keywords: Arbitrary Faults, Distributed And Parallel Computing, Faults, IPPS 2003, Negligible Implementation Cost