/perf-ops

Performance Optimizations and Benchmarks for Distributed SQL Engines

Primary LanguagePythonMIT LicenseMIT

perf-ops

AWS S3 Glue Presto Spark SparkSQL Alluxio Parquet TPC-H

Performance Optimizations Analysis for Distributed SQL Engines

A comparative analysis of Distibuted SQL Engines SparkSQL and Presto

Dataset and Spark ETL

  • Dataset: TPC-H

    TPC-H Schema

    TPC-H Schema

    source

    spark-etl

    ETL Stats

    spark-etl-stats

    7 node: 1 x m4.xlarge(master), 6 x m4.large(worker nodes)

    3 node: 1 x m4.xlarge(master), 2 x m4.large(worker nodes)

Architecture

  • Presto, Spark, Glue, S3

    architecture.png

Results

  • Query Stats

    query-stats

  • Queries

    Q1

    q1

    Q2

    q2

    Q3

    q3

Future Roadmap

  • Automate the cluster provisioning

  • Automate experiments and stats collection

    • Config driven framework

  • Run federated queries

Additional Resources