Performance Evaluation of a Two Level Error Recovery Scheme for Distributed Systems

摘要

Rollback recovery schemes are used in fault-tolerant distributed systems to minimize the computation loss incurred in the presence of failures. One-level recovery schemes do not consider the different types of failures and their relative frequency of occurrence, thereby tolerating all failures with the same overhead. Two-level recovery schemes aim to provide low overhead protection against more probable failures, providing protection against other failures with possible higher overhead. In this paper, we have analyzed a two-level recovery scheme due to Vaidya taking probability of task completion on a system with limited repairs as the performance metric.