Low overhead multiprocessor allocation strategies exploiting system spare capacity for fault detection and location

Several schemes for detecting faults at the processor level in a multiprocessor system have been discussed in the past. One such scheme (A. Dahbura et al., 1989) works by running secondary versions of jobs on the unused or spare processors of the system and uses the comparison approach (J. Maeng and M. Malek, 1981) to detect faults. We build upon this scheme and propose three new multiprocessor allocation strategies that run a variable number of versions per job. These schemes permit online detection and, in many cases, location of faulty processors in a system with nominal degradation in its delay/throughput performance; these delays are limited chiefly to the delays associated with job preemptions. Two new metrics, the fault detection capability (FDC) and the fault location capability (FLC), are introduced to evaluate these schemes. Extensive simulation results are performed to obtain performance figures for the various schemes. Stochastic Petri net models are also developed to obtain approximate performance results. The results show that these schemes utilize spare capacity more efficiently, thereby improving upon the fault detection and location capabilities of the system