Reducing Serialization in Distributed Systems using Mergeable Serialized Replicated Data Types

Abstract: Distributed computing is integral to today’s computing infrastructure. In some cases where high availability is important Conflict-free Replicated Data Types (CRDTs) are used to gain Strong Eventual Consistency between multiple nodes in a network. CRDTs produce a large amount of node-to-node messages which can lead to an overhead from serializing and deserializing messages. This thesis investigates the impact on performance (execution time and memory usage) when performing CRDT operations directly on serialized messages. This thesis studies CRDTs and serialization optimization and techniques. Two types of State-based Convergent Replicated Data Types (CvRDTs) are implemented with operations that operate on serialized data. Microbenchmarks and a larger full-stack experiment are used to measure the effect of using the new operations. The impact on memory usage is inconclusive. A large increase in performance for execution time could be seen in the microbenchmarks. The full-stack experiment showed no impact on performance but had factors that limited how relevant the results were. Operations on serialized data in the context of CRDTs shows promise and warrant further research.