Towards learning a partitioning advisor with deep reinforcement learning

Abstract

In this paper we introduce a partitioning advisor for analytical workloads based on Deep Reinforcement Learning. In contrast to existing approaches for automated partitioning design, an RL agent learns its decisions based on experience by trying out different partitionings and monitoring the rewards for different workloads. In our experimental evaluation with a distributed database and various complex schemata, we show that our learned partitioning advisor is thus not only able to find partitionings that outperform existing approaches for automated data partitioning but is also able to find non-obvious partitionings.