Abstract
Multi-view virtual memory (MVM) provides recoverability for a memory region by defining a view, consisting of access units, on that region. Finite state machines (FSMs) control access to the units of the view. The FSM states are defined to trap on the first write to an access unit in order to save the before-image or to simply change the state of the unit so that an after-image can be recorded at a commit point. The recoverability provided by the MVM approach is compared to other approaches qualitatively, in terms of the interface and functionality, and also quantitatively by examining the delays using simple back of the envelope calculations.