Reliability considerations in large-scale computing systems

The authors address the issue of scalability in the reliability analysis of large-scale degradable homogeneous multiprocessors. The main motivation behind this analysis stems from consideration of whether the reliability of such systems imposes a limit on the number of processors that can cooperate on one problem. Traditional techniques of reliability and performability analysis are used to evaluate the asymptotic behavior of measures such as the mean time to failure and the mission time. The concept of computational reliability is used as a tool to evaluate the measure of reliable computational work as a function of the number of processors. It is shown that the number of processor hours a realistic system can deliver is upper bounded independently from the number of processors. The results demonstrate that graceful degradation in large-scale systems is not scalable; an increase in the number of processors must be matched by a significant increase in the coverage factor in order to maintain the same performance and reliability levels