In this paper, we model the macroscopic behavior of database recovery using
Markov renewal process theory. Specifically, four checkpoint setting schem
es are quantitatively compared based on the reliability evaluation measure
typified by system availability. Finally, numerical examples quantitatively
verify the effects of failures caused by checkpointing and secondary failu
res generated during rollback recovery, in addition to the failures caused
by system failures. (C) 2000 Scripta Technica.