Quorum-based total order broadcast

This project implements a protocol for the coordination of a group of replicas sharing the same mock data. A group of clients sends read and writes requests and all replicas agree by means of a two-phase broadcast. The cluster of replicas is coordinated by a specific replica that determines the order of writes. The protocol implements a ring based election in the event of a crashed coordinator, tolerates multiple replica crashes. Furthermore, the decision of the value is decided using a quorum based approach.

Characteristics

Total order

• Replicas can only commit a value in the order specified by the coordinator.

Virtual Synchrony - Correct epoch delivery

• Messages for which the decision was interrupted due to a coordinator crash are delivered in the current epoch. Instead, messages received during the election of the new coordinator are placed in the subsequent epoch.

Correctness and Liveness

• The commitment of new writes is decided by a quorum of replicas. This ensures the protocol continues even on slow links.
• The election of a new coordinator is ensured to finish using the nodes id as a tie breaker.
• Failed elections restarts if not decided after a fixed amount of time
• Crashed replicas are discovered using time outs.

How to run

Clone the repository, navigate to the root folder of the project and run:

gradle run

A console based menu displays the options to choose from and test the correctness algorithm.