Database Meets AI: A Survey

• Authors: Xuanhe Zhou, Chengliang Chai, Guoliang Li, JI SUN
• Institute: Tsinghua Unversity, Beijing, China
• Published at TKDE'20
• Paper Link: https://ieeexplore.ieee.org/document/9094012

Learning-based Database Configuration

Knob Tuning

Problem: to find the best set of configurations for a DBMS.

Search-based Tuning

Finding the best configurations by branching and bound.

• SoCC'17 - BestConfig: tapping the performance potential of systems via automatic configuration tuning
  • Method
    1. Divides the search space into smaller subspaces
    2. Sampling from the subspaces and iteractively reduces the search space to find the best one
  • Cons
    • Heuristic, no guarantee to find the best one
    • The search space is too large

Traditional ML-based Tuning

Finding the bets configurations using traditional ML-based methods.

• SIGMOD'17 - Automatic Database Management System Tuning ThroughLarge-scale Machine Learning
  • Alias: OutterTune
  • Read Note
• EDBT'19 - SparkTune: tuning Spark SQL through query cost modeling
• Cons
  • The optimal solution obtained in the current stage is not guaranteed to be optimal in other stages.
  • Requires a large number of high quality samples for training.
  • Cannot support too many knobs.

Reinforcement Learning for Tunning

Uses a Reinforcement Learning (RL) agent to find the best configurations for a DBMS.

• SIGMOD'19 - An End-to-End Automatic Cloud Database Tuning System Using Deep Reinforcement Learning
  • Alias: CDBTune
  • Method
    • The RL Modeling:
      • Environment: a cloud DBMS
      • State: the internal metrics of the DBMS (similar to OutterTune)
      • Action: the values for increasing or decreasing configurations (knobs)
      • Reward: the difference of DBMS's performance
      • Agent Model: Deep Deterministic Policy Gradient (DDPG)
  • Pros
    • Does not need high-quality training data
  • Cons
    • without considering workload features
• VLDB'19 - QTune: A Query-Aware Database Tuning System with DeepReinforcement Learning
  • Alias: QTune
  • Method
    • Basically same with CDBTune but considers workloads.
    • Uses Double-state Deep Reinforcement Learning (DS-DRL)

(Reading...)