Performance

Comparing the Performances of Swift SQLite libraries

Last updated May 2, 2020

Below are performance benchmarks made on for GRDB 5.0.0-beta, FMDB 2.7.5, and SQLite.swift 0.12.2. They are compared to Core Data, Realm 4.4.0 and the raw use of the SQLite C API from Swift.

This report was generated on a MacBook Pro (15-inch, Late 2016), with Xcode 11.4.1, by running the following command:

make test_performance | Tests/parsePerformanceTests.rb | Tests/generatePerformanceReport.rb

All tests use the default settings of each library. For each library, we:

• Build and consume database rows with raw SQL and column indexes (aiming at the best performance)
• Build and consume database rows with column names (sacrificing performance for maintainability)
• Build and consume records values to and from database rows (aiming at the shortest code from database to records)
• Build and consume records values to and from database rows, with help from the Codable standard protocol
• Build and consume records values to and from database rows, with change tracking (records know if they have unsaved changes)

As a bottom line, the raw SQLite C API is used as efficiently as possible, without any error checking.

	GRDB	Raw SQLite	FMDB	SQLite.swift	Core Data	Realm
Column indexes
Fetch	0.09	0.06	0.06	0.55	¹	¹
Insert	0.10	0.02	0.16	0.12	¹	¹
Column names
Fetch	0.16	¹	0.23	0.88	¹	¹
Insert	0.13	¹	1.02	0.86	¹	¹
Records
Fetch	0.17	0.06	2.86	0.85	¹	¹
Insert	0.27	¹	¹	¹	¹	¹
Codable Records
Fetch	0.32	¹	¹	¹	¹	¹
Insert	0.25	¹	¹	¹	¹	¹
Records with change tracking
Fetch	0.41	¹	¹	¹	0.34	0.16
Insert	0.32	¹	¹	¹	0.39	0.58

¹ Not applicable

• Column indexes:

  • Fetch (source)

    This test fetches 100000 rows of 10 ints and extracts each int given its position in the row.

    It uses FMDB's -[FMResultSet longForColumnIndex:], GRDB's Row.value(atIndex:), and the low-level SQL API of SQLite.swift.

  • Insert (source)

    This test inserts 20000 rows of 10 ints, by setting query arguments given their position.

    It uses FMDB's -[FMDatabase executeUpdate:withArgumentsInArray:] with statement caching, GRDB's UpdateStatement.execute(arguments:Array), and the low-level SQL API of SQLite.swift.

• Column names:

  • Fetch (source)

    This test fetches 100000 rows of 10 ints and extracts each int given its column name.

    It uses FMDB's -[FMResultSet longForColumn:], GRDB's Row.value(named:), and the high-level query builder of SQLite.swift.

  • Insert (source)

    This test inserts 20000 rows of 10 ints, by setting query arguments given their argument name.

    It uses FMDB's -[FMDatabase executeUpdate:withParameterDictionary:] with statement caching, GRDB's UpdateStatement.execute(arguments:Dictionary), and the high-level query builder of SQLite.swift.

• Records:

  • Fetch (source)

    This test fetches an array of 100000 record objects initiated from rows of 10 ints.

    It builds records from FMDB's -[FMResultSet resultDictionary], GRDB's built-in FetchableRecord protocol, and the values returned by the high-level query builder of SQLite.swift.

  • Insert (source)

    This tests inserts 20000 records with the persistence method provided by GRDB's PersistableRecord protocol.

• Codable Records:

  • Fetch (source)

    This test fetches an array of 100000 record objects initiated from rows of 10 ints.

    It builds records from GRDB's built-in support for the Decodable standard protocols.

  • Insert (source)

    This tests inserts 20000 records with the persistence method provided by GRDB's built-in support for the Encodable standard protocols.

• Records with change tracking:

  • Fetch (source)

    This test fetches an array of 100000 record objects initiated from rows of 10 ints.

    It builds records from FMDB's -[FMResultSet resultDictionary], GRDB's built-in Record class.

  • Insert (source)

    This tests inserts 20000 records with the persistence method provided by GRDB's Record class.