Distributed Consensus Algorithms to Improve Database Consistency: Optimization of Three-Phase Commit Protocol

Abstract

In computer networking and databases, the two-phase commit and three- phase commit protocols are distributed algorithms that make all nodes in a distributed system agree to either commit or abort a transaction. In both the protocols, one node is designated the coordinator, which is the master site, and the rest of the nodes in the network are designated the cohorts.

The 2PC algorithm is a blocking protocol, in which, a node will block while it is waiting for a message. This has the effect that resources are tied up forever. The 3PC protocol, on the other hand, eliminates blocking by using the failover mechanism; in which the failed coordinator is replaced by an elected cohort and the failed cohort is simply discarded from the commit process. The disadvantage of the 3PC protocol is the higher complexity of the protocol in terms of time and communication, which renders it infeasible to use in the real world applications.

As a result, 2PC is employed by either loosening the consistency constraints or by using hardware backup mechanisms to handle its blocking nature; which is not quite affordable. Thus, ultimately, the challenge lies in optimizing the 3PC protocol so that the time and communication complexities can be reduced while keeping the consistency intact.

Hence, the objective of this dissertation work is to identify the redundancies involved in the protocol mechanism and modify the algorithm in order to eliminate them so as to reduce the protocol complexity. The goal is to make the 3PC viable to be used in the real world transaction management systems. In this project, two optimization methods have been used – one is the Backup Node Optimization, and second is the Fast Convergence Method.