A Temporal Consensus Model

Abstract

The active (state-machine) replication protocol has been one of the commonly adopted approaches to provide fault tolerant data services. In such a scheme, service state is replicated at all participating servers and client requests are presented to all the servers. A strict consensus protocol executed by the servers ensures that all servers receive all the requests in the same order. The output of each request from all the servers are then compared to mask the result of faulty server(s). However, the overhead associated with running the strict consensus protocol tends to slow down the response to client requests, which makes it unsuitable for real-time applications where timing predictability is a necessity. This paper presents a weak consensus model called “temporal consensus” that reduces such overhead and enable the active scheme to meet the timing constraint of real-time applications.