Reliability and fault-tolerant issues of multiprocessor and multicomputer systems

This paper deals with the reliability and fault-tolerance evaluation of multiprocessor and multicomputer architectures considering the degradation of both computation and communication capabilities. Reliability and performance availability (pa) are used to characterize and evaluate the dependability of these architectures. Bandwidth availability (ba) and computation-communication availability (cca) are used to quantify the pa of multiprocessors and multicomputers, respectively. These measures are based on the system requirements for the parallel execution of a task (job) that consists of a few subtasks. We present two different dependability models for multiprocessors, namely: a bus-oriented model (bom) and a switch-oriented model (som). The bom is an analytical model and is used to evaluate multiprocessors with crossbar and multiple-bus interconnections. The som uses simulation to analyze all types of multiprocessors. A simulation technique is also presented to compute the reliability and cca of various types of multicomputer networks suggested in the literature.