Design of Database Index Structure based on Optimized Deep Reinforcement Learning

Databases are necessary for the growth of the cloud computing to efficiently manage large amounts of data. For real-time data processing in Not only Structured Query Language (NoSQL) databases, effective index services are critical and when it comes to enhancing NoSQL database performance, index selection is crucial. A Deep Reinforcement Learning Index Selection Method (DRLISM) is used for choosing indexes in NoSQL databases. To maximize database performance, various indexes and their various characteristics are selected for various workloads. In order to determine an optimum index for specific workload as well as adapt to a dynamic workload, the DRLISM builds on the Modified Beetle Antenna Optimisation with Sparrow Search Optimisation (DRLISM-MBAOSSO). According to standard single index structures, experimental data indicate that DRLISM has better performance in terms of increased throughput to varied degrees. Similarly, when compared with conventional approaches such as: Integrating Rules and Reinforcement Learning to Improve Index Recommendation (RBOIRA) and Deep Reinforcement Learning (DRL), the proposed DRLISM-MBAOSSO achieved a higher throughput of 99.73%.