A phase expansion for non-Markovian availability models with time-based aperiodic rejuvenation and checkpointing. Issue 15 (2nd August 2020)

• Record Type: Journal Article
• Authors: Zheng, Junjun; Okamura, Hiroyuki; Dohi, Tadashi
• Is Part Of: Communications in statistics. Volume 49:Issue 15(2020)
• Page Start: 3712
• Page End: 3729
• Publication Date: 2020-08-02
• Subjects: Software rejuvenation -- checkpointing -- stochastic reward net -- non-Markovian process -- phase expansion -- optimal rejuvenation trigger timing -- steady-state availability; Mathematical statistics -- Periodicals; Mathematics; Statistics; 519.2
• DOI: 10.1080/03610926.2019.1708400
• Languages: English
• ISSNs: 0361-0926

Abstract

Abstract: This paper presents a stochastic framework, consisting of stochastic reward net (SRN) for capturing the transient behaviors of the system and its related non-Markovian state transition diagram, to model an operational software system that undergoes aperiodic time-based rejuvenation and checkpointing schemes, and further to investigate whether there exists the optimal rejuvenation schedule that maximizes the system steady-state availability. A phase expansion approach is adopted to solve the non-Markovian availability models, which are actually neither the semi-Markov processes nor the Markov regenerative processes. Our numerical results show an appropriate rejuvenation trigger timing range, resulting in the positive improvement effect on the system availability of a database system, and that there exists the optimal rejuvenation trigger timing maximizing the system availability.