COMPARATIVE-ANALYSIS OF DIFFERENT MODELS OF CHECKPOINTING AND RECOVERY

Checkpointing and rollback-recovery is a common technique to keep the integrity of information and to enhance the reliability in database systems. Several models were studied in order to determine the optimum checkpointing policy which optimizes a certain performance measure. Among checkpointing strategies that have been considered are: 1) Poisson checkpointing, 2) a fixed time interval between checkpoints and 3) a specified number of completed transactions between checkpoints. Poisson checkpointing is often assumed because it is more tractable analytically. The second and third strategies are more realistic, but also more difficult to analyze. An important aspect of checkpointing and recovery models is the dependence of recovery periods on the checkpointing strategy. This dependence may be characterized as one of the following: 1) parametric dependence, 2) stochastic dependence and 3) deterministic dependence. Parametric dependence is the least realistic and simplest to model. Naturally, more realistic models are more complex to analyze. The purpose of this paper is to study and compare different models in order to select one which adequately represents a realistic system and yet tractable for analysis. The approach is analytical (whenever possible), otherwise numerical or by means of simulations. In particular, we consider the queueing analysis of a model with combined checkpointing strategies; the results can be specialized to any one of the combined strategies. © 1990 IEEE