INDEX RECOMMENDATION BASED ON DUAL CHANNEL DQN

Aiming at the problems such as unused the characteristics of SQL workload and the mechanical rule method during index recommendation, we propose a DQN-based dual channel index recommendation model (Dual Channel Deep Q-Network, DC-DQN). The index selectivity and SQL query type features wre trained independently through two separate channels and the information fusion was carried out through the full connection layer, so as to select the candidate index that better matches three-star index. The experimental results on TPC-H dataset show that DC-DQN performs as good as having all indexes and under the construction of specific query workload, DC-DQN performs better than the previous method.