From Serializable to Causal Transactions for Collaborative Applications

Overview & Synopsis

Serializability is the traditional consistency criterion when shared objects are accessed concurrently. Its main drawback lies in the strong synchronization constraints it imposes on execution of applications when they are run on distributed systems. In this paper we examine weake consistency criteria for computations in which accesses to shared objects are grouped to form transactions. In particular, we explore causal consistency and causal serializability for transaction based computations. These criteria turn out to be sufficient for a class of applications ( e.g., collaborative applications) and their implementation results in greater availability of data and improved performance. These criteria are formally defined and protocols implementing them are presented. We demonstrate that causal consistency allows both read and update transactions to be executed in a wait-free manner. Although write accesses in causal serializability require synchronization with other nodes, read accesses appearing in update transactions can be executed locally. This distinguishes causal serializability from serializability where synchronization is necessary for both read and write accesses of update transactions. Finally, fault-tolerance problems are addressed in the context of process crash failures.