Proving Conflict Serializability for Full Invalidation

To address the need for improved efficiency and to support an ever-increasing feature set, the specification and implementation of state-of-the-art transactional memory (TM) systems are vastly different than the pioneering systems from a decade ago. An unfortunate side-effect of this shift is that some of the early theoretical work on TM correctness does not apply to these new systems. This paper presents a work in progress to show a full invalidation automaton only accepts schedules that are conflict serializable, a principle correctness criterion for TM. In this paper we present the formalization of a full invalidation automaton and the general direction of our approach. We are interested in proving full invalidation correct because it is a recently devised method for performing transactional opacity and conflict detection and has been shown to be efficient for InvalSTM, a software-only TM (STM) that is competitive with other state-of-the-art STMs.