Skip to content
This repository was archived by the owner on Jan 10, 2023. It is now read-only.

Files

Latest commit

 

History

History

perf

README.md

Lovefield Performance Benchmark

Tests

The following categories of benchmark tests are orchestrated.

  • Loading: measures time of loading an empty DB and a populated DB (with 20K rows).

  • Full table SCUD: perform bulk insert/update/select/delete on 10K/20K/30K/40K/50K rows for both IndexedDB (Full table SCUD) and MemDB (Full table SCUD Mem). The queries performed are guaranteed to be executed by a plan of full table scan.

  • PK-based SCUD: similar to full table SCUD, but this time a primary key (i.e. unique B+-Tree index) will be involved in all SCUD operations.

  • Select: exhaustively test different select scenarios for both IndexedDB and MemDB:

    • Select single row with predicates
      • Indexed
      • Partially indexed (i.e. using only one key for a two-key index)
      • Non-indexed
    • Select many rows with predicates
      • Indexed range scan for continuous range
      • Indexed "spaced out", means the column to be searched is indexed, but non-continuous. For example, pick 1, 10, 20, 30 from an indexed column.
      • Non-indexed
      • Non-indexed spaced out
    • Select all with projections
      • Select all
      • Select aggregations, and the aggregated column is indexed.
      • Select aggregations, but the aggregated column is not indexed.
    • Select all with ordering
      • The ordered by column is indexed.
      • The ordered by column is partially indexed.
      • The ordered by column is not indexed.
      • When limit and skip is present and the ordered by column is indexed.
    • Select all with join operation
      • "Equi": simple inner join, WHERE e.jobId == j.id
      • "Theta": combined search condition, WHERE e.jobId == j.id AND e.salary > j.maxSalary

  • Scenario: simulates one scenario of real-world usage, MemDB only

    • Insertion of 1K rows via tx.attach()
    • Select 1K rows via 1K parallel queries
    • Select 1K rows via 1K parallel queries, reusing the query builder via parameterized query.

Testing Configurations

The tests are run on an Ubuntu workstation with Dual Xeon 2690 2.9GHz, 64GB RAM, 7200rpm SATA-II drive. Tests are performed Monday to Friday on latest Chrome browser stable version.

The numbers can be queried using the dashboard provided in the source code.