Evaluation of Eventual Consistency and Linearizability in MongoDB

Abstract

Sharding and replication are crucial techniques for scaling distributed systems,enabling data distribution across multiple nodes. However, as the number ofreplicas increases, maintaining consistency across them becomes increasingly challenging.Developers require understanding the consistency guarantees offered bydifferent distributed systems to make informed decisions about the trade-offs betweenconsistency and low latency. The study of consistency is essential to maintaindata integrity and optimize performance and scaling.In this work, we focus on evaluating the eventual consistency and linearizabilityprovided by MongoDB, a popular distributed database system. The experimentconsiders various combinations of read and write concern levels and howthey affect the consistency of the system. We also take into account different sizesof the document as a parameter in measuring the consistency. By analyzing thesefactors, we aim to quantify the impact of different parameters on the system�sconsistency.We evaluate the performance of MongoDB by measuring the read and writelatency of the operations by varying the read and write concern levels as well asby varying the document size. The evaluation of linearizability is based on measuringthe occurrence of stale reads, that happen when a read operation accessesoutdated or inconsistent data. By analyzing these variables, the aim is to providea more comprehensive understanding of MongoDB�s eventual consistency andlinearizability and how it behaves in different scenarios.Our findings reveal that using "linearizable" readConcern has a significantimpact on the read latency and hence should only be used in scenarios wherestrong consistency guarantee is absolutely essential. Furthermore, document sizehas a significant impact on write latency and consistency but not on the read latency.Using "majority" readConcern and writeConcern provides a good balancebetween consistency and latency in MongoDB.