Availability and performability analysis for a service degradation process with condition-based preventive maintenance I - formulation and optimisation

Abstract

The preventive maintenance is very useful to improve effectively the service availability for software systems with service degradation. In this paper, we present a stochastic model to describe an operational software system, which consists of one operating system and multiple applications and provides a service in continuous time. Two kinds of rejuvenation strategies are taken, namely reconfiguration of applications as a corrective maintenance and reinstallation of an operating system as a preventive maintenance. We derive the optimal preventive rejuvenation schedules maximising the steady-state service availability and maximising the expected reward per unit time, by means of semi-Markov decision processes. Illustrative numerical examples are presented to give decision tables on the optimal software rejuvenation policies.

Availability and performability analysis for a service degradation process with condition-based preventive maintenance II - estimation and simulation

“…To complete the discussion, we consider a software system which consists of one operating system and multiple applications in Dohi (2013). The system can provide a service in continuous time and, at the same time, may deteriorate with time.…”

“…Suppose that the service level of software at time t is described by a semi-Markov process (SMP) with 2(s + 1) states and constant sojourn time k j (j = 1, • • • , s). As Dohi (2013) show, since this SMP can be simplified to a right-skip free CTMC by aggregation of states…”

“…Hence, the notions of service reliability and service availability are becoming popular in our highly information-oriented society. In Dohi (2013), we develop a dynamic software rejuvenation model and derive the optimal software rejuvenation policy, where the transition law of the service state is assumed to be known. In general, the model parameters, which are involved in stochastic models, need to be estimated from the underlying data.…”

“…However, this is not trivial for the operational service systems because it is difficult to observe a sufficient number of data before the service operation. Dohi (2013) consider the Markov and semi-Markov decision process models to trigger the software rejuvenation policy by monitoring the system degradation state, where the underlying transition rates have to be known. In other words, the service degradation process should be estimated in any adaptive wayas the service time goes on.…”

“…According to Okamura et al (2010) and Okamura and Dohi (2011), we also develop the Q-learning algorithm for the service availability maximisation and the expected reward minimisation problems in Dohi (2013). The most attractive point is that the Q-learning algorithm is statistically consistent, i.e., the estimate of the optimal software rejuvenation time approaches to the real (but unknown) optimal solution as the number of observations increases.…”