On the Analysis of Software Rejuvenation Policies

Software rejuvenation is a technique for software fault tolerance which involves occasionally stopping the executing software, \cleaning" the \internal state" and restarting. This cleanup is done at desirable times during execution on a preventive basis so that unplanned failures, which result in higher costs compared to planned stopping, are avoided. Since during rejuvenation, the software is typically unavailable or in a degraded mode of operation, the operation involves a cost. The necessity to use this technique not only in general purpose computers but also in safety-critical and high availability systems clearly indicates the need of analysis in order to determine the optimal times to rejuvenate. In this paper, we present an analytical model of a software system which services transactions. Due to \aging", not only the service rate of the software decreases with time but the software itself experiences occasional crash/hang failures. We propose and compare two rejuvenation policies. First policy is purely time dependent while the second also takes into account the number of transactions currently queued for service. The policies are evaluated for the resulting steady state availability as well the probability that a transaction is denied service. We also numerically illustrate the use of our model to compute the optimal rejuvenation interval which manimizes (maximizes) the loss probability (steady state availability).