Modeling and evaluation of relational data concepts on top Redis in-memory key-value storage

Abstract

This thesis investigates the ability of creating relational data concept on top of NoSQL data store. Implementation of concept includes using of consistency key naming and in-memory Redis structures for representation of regular relational entities. Search mechanism in this concept was implemented using sets of index values and predefined key names, which is providing high speed of read operations. The main objective was to model the system that could replicate relational queries without inheriting the complexity of regular relational models using SQL engine. Then evaluate this system. To this end, an automated information system was developed that loaded data into a key-value store by generating pre-calculated access keys for entries and writing them in the required structures. After loading, same system can executed SQL queries, the primary search mechanism of which was based on calculating the access key through index sets. To evaluate the data storage architecture, a synthetic dataset was generated, and tests of various operations were conducted to collect metrics for further efficiency and performance evaluation. A comparative analysis with other relational data storage systems was also conducted. As a result, a model was created that implements relational abstraction without inheriting architectural complexity. The developed model can execute SQL queries and demonstrates high performance for point search operations.