Skip to content

PERFORMANCE.md

Latest commit

 

History

History
32 lines (26 loc) · 1.48 KB

PERFORMANCE.md

File metadata and controls

32 lines (26 loc) · 1.48 KB
Raw

Performance and profiling

The provider has two different cost centres and they must be measured separately:

  • the file layer (page decoding, row writes and index maintenance), and
  • the ADO.NET mirror (one-time SQLite materialization versus repeated query execution).

The opt-in InsertBenchmarkTests test measures 100,000 low-level rows with the same schema in managed C# and Jackcess 5.1.5. It reports insert/read milliseconds, rows/second and the cost of the non-atomic WriteBatch path. The benchmark is deliberately excluded from normal CI because timings depend on the host and Java installation:

$env:UCANACCESS_PERF = '1'
$env:UCANACCESS_PERF_ROWS = '100000'
dotnet test tests/UCanAccess.Tests/UCanAccess.Tests.csproj -c Release `
  --filter FullyQualifiedName~InsertBenchmarkTests --logger 'console;verbosity=normal'

For mirror profiling, run the regular provider tests under a sampling profiler (PerfView, dotnet-trace or Visual Studio) and compare Keep Mirror=true with Mirror Mode=file. The useful boundaries are Mirror.BuildSchemaAndLoad, Mirror.LoadData, AccessSqlTranslator.Translate and UCanAccessConnection.ExecuteDmlBatchAtomically. File-backed mirrors are a cache and are rebuilt at open; they are not a database transaction log.

When changing page traversal, row codecs or mirror loading, record the row count, database size, runtime/OS and both C# modes in the pull request. A benchmark result without those inputs is not comparable across machines.