Evolutionary design of fault tolerant collective communications

Abstract

Scheduling of collective communications (CC) in interconnection networks possibly containing faulty links has been done with the use of the evolutionary techniques. Inter-node communication patterns scheduled in the minimum number of time slots have been obtained. The results show that evolutionary techniques often lead to ultimate scheduling of CC that reaches theoretical bounds on the number of steps. Analysis of fault tolerance by the same techniques revealed graceful CC performance degradation for a single link or node fault. Once the faulty region is located, CC can be re-scheduled during a recovery period. © 2008 Springer-Verlag Berlin Heidelberg.