Dynamic analysis of the relay cache-coherence protocol for distributed transactional memory

Transactional memory is an alternative programming model for managing contention in accessing shared in-memory data objects. Distributed transactional memory (TM) promises to alleviate difficulties with lock-based (distributed) synchronization and object performance bottlenecks in distributed systems. In distributed TM systems, both the management and consistency of a distributed transactional object are ensured by a cache-coherence protocol. The Relay protocol is a cache-coherence protocol that operates on a fixed spanning tree. The protocol efficiently reduces the total number of abortions for a given set of transactions. We analyze the Relay protocol for a set of transactions which are dynamically generated in a given time period, and compare the protocol's time complexity against that of an optimal offline clairvoyant algorithm. We show that Relay is O(log D)-competitive, where D is the diameter of the spanning tree, for a set of transactions that request the same object, given the condition that the maximum local execution time of transactions is sufficiently small.