A Case of Multi-Level Distributed Recovery Schemes
Nitin H. Vaidya · 2001
Most of the distributed recovery schemes proposed in the literature are designed to tolerate arbitrary number of failures, with a few notable exceptions of schemes designed to tolerate single failures. In this report, we demonstrate that, it is often advantageous to use "multi-level" recovery schemes. A "multi-level" recovery scheme is one that can tolerate different number of faults at different costs, tolerance of larger number of failures requiring larger costs. The costs are incurred during failure-free operation as well as during recovery. To demonstrate the advantages of multi-level recovery, we analyze a hypothetical 2-level recovery scheme that takes two different types of checkpoints, namely, 1checkpoints and N-checkpoints. A single failure can be tolerated by rolling the system back to a 1-checkpoint, while multiple failure recovery is possible by rolling back to an N-checkpoint. The cost of a 1-checkpoint may be expected to be smaller than that of an N-checkpoint. For such a...