Index Advisor based on Deep Reinforcement Learning

October 31, 2020 ยท View on GitHub

Code for CIKM2020 paper

What does it do?

This is an index advisor tool to recommend an index configuration for a certain workload under maximum storage or index number. It combines the heuristic rules and deep reinforcement learning together.

What do I need to run it?

  1. You should install a PostgreSQL database instance with HypoPG extension.
  2. You should install the required python packages (see environment.yaml exported from conda).
  3. In this code, we adopt TPC-H. Thus, you construct your own TPC-H database instance.
  4. We need the TPC-H tool to generate the workload. You can download it from this page.

How do I run it?

  1. You can find the entry in Entry/EntryM3DP.py
  2. There is a sample about how to use the workload and index candidates generation algorithms in Utility/Sample4GenCandidates.py.

Notice

  1. The index candidates generated algorithms (parser and generation algorithms in Utility/ParserForIndex.py) are for TPC-H cases. It may be not suitable TPC-DS. Because some query patterns are not in TPC-H.