A comparison of four adaptation algorithms for increasing the reliability of real-time software

Abstract

In a large, parallel, real-time computer system, it is frequently most cost-effective to use different software reliability techniques (e.g., retry, replication, and resource-allocation algorithms) at different levels of granularity or within different subsystems, dependent on the reliability requirements and fault models associated with each subsystem. The authors describe the results of applying four reliability algorithms, via RESAS (REal-time Software Adaptation System), to a sample real-time program executing on a multiprocessor.<>