Unlocking the Power of Diversity in Index Tuning for Cluster Databases

Haitian Hang, Xiu Tang, Bo Zhou, Jianling Sun

Abstract: Index tuning is crucial for database performance, but existing algorithms are often tailored for single instances, posing challenges in cluster architectures. While uniform tuning simplifies matters by applying identical configurations across nodes, it overlooks the advantages of diverse configurations in clusters. Conversely, heterogeneous tuning, accommodating varied configurations across instances, optimally utilizes cluster resources. This paper introduces a practical framework for such tuning, compatible with existing tools. Our workload-driven approach tailors optimization for each instance, complemented by a benefit-first load-aware routing strategy to enhance performance. Extensive experiments using PostgreSQL and TPC-H/TPC-DS benchmarks validate the efficacy of our methods.