Consistency-Latency Trade-Off of the LibRe Protocol: A Detailed Study

In multi-writer, multi-reader systems, data consistency is ensured by the number of replica nodes contacted during read and write operations. Contacting a sufficient number of nodes in order to ensure data consistency comes with a communication cost and a risk to data availability. In this paper, we extend our previous work on a consistency protocol called LibRe, which helps to read the latest version of a data item by contacting a minimum number of replica nodes. The protocol uses a registry that records the set of replica nodes containing the most recent version of the data items until all replicas of this data item converge to a consistent state. Hence, referring to the registry during read time helps to forward the read requests to the replica nodes holding the most recent version of the needed data item. In the following work, we show that this protocol provides a new trade-off between consistency and latency for distributed data storage systems. We provide a formal description of the protocol and its reliability and evaluate the scalability of the protocol up to one hundred nodes by simulation. We also demonstrate the effectiveness of the approach in practice by providing a proof-of-concept implementation of the protocol inside the Cassandra distributed data store. The test results prove that using LibRe protocol, an application would experience a similar number of stale reads compared to strong consistency options offered by Cassandra, while achieving lower latency, and similar availability.