Towards learning a partitioning advisor with deep reinforcement learning

Abstract

Commercial data analytics products such as Microsoft Azure SQL Data Warehouse or Amazon Redshift provide ready-touse scale-out database solutions for OLAP-style workloads in the cloud. While the provisioning of a database cluster is usually fully automated by cloud providers, customers typically still have to make important design decisions which were traditionally made by the database administrator such as selecting the partitioning schemes. In this paper we introduce a learned partitioning advisor for analytical OLAP-style workloads based on Deep Reinforcement Learning (DRL). The main idea is that a DRL agent learns its decisions based on experience by monitoring the rewards for different workloads and partitioning schemes. We evaluate our learned partitioning advisor in an experimental evaluation with different databases schemata and workloads of varying complexity. In the evaluation, we show that our advisor is not only able to find partitionings that outperform existing approaches for automated partitioning design but that it also can easily adjust to different deployments. This is especially important in cloud setups where customers can easily migrate their cluster to a new set of (virtual) machines.

External vs. Internal: An Essay on Machine Learning Agents for Autonomous Database Management Systems

“…Given the general-purpose nature of RL, it is one of the most active areas of DBMS tuning research in the late 2010s. Researchers have applied RL for query optimization [7], index selection [10], partitioning [3,6].…”

UDO: Universal Database Optimization using Reinforcement Learning

“…Recently, there has been significant interest in using machine learning for database tuning [12,19,20,25,27]. Our work falls into the same, broad category as it exploits RL.…”

Qd-tree: Learning Data Layouts for Big Data Analytics

“…5 has filters on supplier's r_name, a categorical with diverse literalsqd-tree yields a 16.8× speedup on this template. 19 is an OR of three complex 6-filter blocks; qd-tree is able to optimize for this complex template and provides 5.5× speedup over Bottom-Up. Bottom-Up is faster only on 1 and 18 , both of which require the full month worth of data.…”

“…This line of research leverages recent advancements in deep learning algorithms and scalable hardware (GPU) to improve database systems. Closest to our work is a proposal of learned partition adviser using deep RL [19]; it focuses on replication and coarse-grained partitioning (e.g., hash) along entire attribute(s), unlike qd-tree which partitions based on a rich set of fine-grained candidate cuts. In this space, machine learning has also been used to revisit tuning [47], workload forecasting [28], data structures and indexes [12,20,23,32], and query optimization [13,24,29,50].…”