Liveness in Transactional Memory

Abstract

In this chapter we give a formal overview of liveness properties of transactional memory (TM) systems. Unlike safety properties, which require some ‘bad’ events not to occur, liveness properties require some ‘good’ events to eventually occur. Usually, liveness properties of shared memory systems require some operations to eventually return a response (terminate). However, in the context of TM systems operation termination is not enough to ensure meaningful progress. It is necessary to require some transactions to eventually commit. In this chapter we give precise definitions of liveness properties and what it means for a TM systems to satisfy a liveness property. Using the defined formal framework we give some impossibility results. We show that it is impossible to guarantee both local progress, the strongest TM liveness property that requires every correct transaction to eventually commit, and common TM safety properties such as strict serializability or opacity in a fault prone system.