Analysis of Query Optimization Using Deep Reinforcement Learning Using Particle Swarm Optimization Algorithms

Query optimization is a well-studied problem in the database industry, with numerous solutions proposed over the last several decades. The success of deep reinforcement learning (DRL) has generated new opportunities in query optimization. One of the most difficult tasks in query optimization and query plan generation is determining the order in which join operations between tables are done (i.e. relations). Even if the final results of a query remain identical regardless of join order, the order in which the tables of a query are joined can have a significant impact on query execution time. Deep reinforcement learning, in particular a data-driven method to reasoning about enumeration heuristics, provides a novel algorithmic viewpoint on join enumeration. We must now control what training data the model views and how that data is featured, rather than the standard tunable parameters of a query optimizer. The algorithm makes few assertions about the cost model's structure or the search space's topology. We demonstrate that Q learning optimizes plans well across many different cost models for a small set of training queries. On the TPC-H database, the Query Optimization Algorithms Q-Learning and PSO (Particle Swarm Optimization) are assessed. During the evaluation, the optimizer failed to complete one query within the maximum time permitted, whereas the deep reinforcement learning-based models (Q-Learning) and heuristics model (PSO) managed. Of course, the standard join ordering problem is NP-hard, and practical algorithms use heuristics to make the search for a good plan efficient. A novel method for Query Optimization using Particle Swarm Optimization (QOPSO) and Deep Q Learning (DQL) for parameter tuning of Join Operation Cost and Processing Cost is suggested in this paper.