Peleton DB

Introduction

  • Challenges

    • understand an application's workload, OLTP or OLAP.
    • optimization decisions rely on the above.
    • needs to forecast resource utilization trends

  • Areas to Apply

  • Physical

    • Indexes
    • Materialized Views
    • Storage Layout

  • Data

    • Location
    • Partitioning

  • Runtime

    • Resources
    • Configuration Tuning
    • Query Optimizations

  • For each optimization action, DBMS needs to estimate effect on DB.

  • Optimizations need to be applied in such a way that there is no perceptible impact to the application during deployment.

  • Latency is the most important metric in a DBMS as it captures all aspects of performance.

  • Components

    • Workload classification
      • clustering the workload
        • use runtime metrics
        • query logical semantics
    • Workload Forecasting
      • Auto-regressive-moving average model
    • Action Planning and Execution
      • Advantage of having tight coupling of the autonomous components and DBMS architecture is most evident, as they can provide feedback to each other.
      • Action generation
      • Action planning
        • Receding horizon-control model.
      • Deployment

  • Additional Considerations

    • Explanation of chosen action plans.
    • Priotity across components

  • See Noise Page