OPCAM: Optimal Algorithms Implementing Causal Memories in Shared Memory Systems

Abstract

Data replication is commonly used for fault tolerance in reliable distributed systems. In this paper, we propose three optimal protocols for causal consistency in distributed shared memory systems. Our proposed optimal protocols are designed for partial replication across the distributed shared memory. Complete replication is a special case of our protocols and we also give the optimal implementation of causal consistency for the complete replication case. Algorithm Full-Track is optimal in the sense that it can update the local copy as soon as possible while respecting causal consistency. Algorithm Opt-Track is further optimal in the sense that the size of the local logs maintained and the amount of control information piggybacked on the update messages is minimal. Algorithm Opt-Track-CRP is a special case of algorithm Opt-Track for the full replication case. It is highly scalable, and significantly more efficient than the Baldoni et al. protocol for the complete replication case.