UA-.NETStandard/tests/Opc.Ua.PubSub.Tests/Benchmarks/README.md

Opc.Ua.PubSub.Tests benchmarks

BenchmarkDotNet suite covering the four hot paths of the Part 14 v1.05.06 PubSub stack: UADP encode/decode, JSON encode/decode, scheduler tick dispatch, and AES-128-CTR sign+encrypt.

Quick smoke pass

The dry-run smoke pass takes ~10 seconds, runs every benchmark exactly once, and emits a summary table that can be diffed for catastrophic regressions:

dotnet run -c Release -p tests\Opc.Ua.PubSub.Tests\Opc.Ua.PubSub.Tests.csproj `
  -f net10.0 -- --job dry --filter '*' --inProcess

The reference output for the most recent commit is checked in at Baselines/baseline-net10-dry.md.

The --inProcess flag forces InProcessEmitToolchain. Without it BenchmarkDotNet generates a satellite project that doesn't honour our solution's Directory.Build.props and Directory.Build.targets, so the source generators don't run (MODELGEN003). The in-process toolchain runs benchmarks in the BDN host process and is the only toolchain that works without bespoke BDN configuration.

Real benchmark runs

--job dry is not statistically valid (one warm-up + one iteration). For real numbers use one of the longer jobs:

# ~5 minutes total. Single launch, ~3 iterations per benchmark.
dotnet run -c Release -p tests\Opc.Ua.PubSub.Tests\Opc.Ua.PubSub.Tests.csproj `
  -f net10.0 -- --job short --filter '*' --inProcess

# ~30 minutes total. Multiple launches, ~15 iterations each.
dotnet run -c Release -p tests\Opc.Ua.PubSub.Tests\Opc.Ua.PubSub.Tests.csproj `
  -f net10.0 -- --job medium --filter '*' --inProcess

# ~3 hours total. The defaults — full statistical pipeline.
dotnet run -c Release -p tests\Opc.Ua.PubSub.Tests\Opc.Ua.PubSub.Tests.csproj `
  -f net10.0 -- --filter '*' --inProcess

Filter to one suite to iterate locally:

dotnet run -c Release -p tests\Opc.Ua.PubSub.Tests\Opc.Ua.PubSub.Tests.csproj `
  -f net10.0 -- --filter '*UadpEncoding*' --inProcess

Output lands under BenchmarkDotNet.Artifacts/results/ next to the project. To save outside the repo:

dotnet run -c Release -p tests\Opc.Ua.PubSub.Tests\Opc.Ua.PubSub.Tests.csproj `
  -f net10.0 -- --filter '*' --inProcess `
  --artifacts $env:USERPROFILE\bench-results

Baselines

Baselines/ holds the smoke-pass summary tables that this commit was verified against.

• baseline-net10-dry.md — dry job on net10.0.

When a hot-path change is intentional, regenerate the baseline by re-running the smoke pass and committing the updated table in the same PR.

Suites

• UadpEncodingBenchmarks — UADP EncodeAsync / TryDecodeAsync across SingleField (UInt32), TenFields (mixed primitives), HundredFields (mixed primitives), Strings (10×64 char fields), LargeArray (Float[256]).
• JsonEncodingBenchmarks — Same dataset shapes, two encoder modes (Verbose, Compact).
• SchedulerBenchmarks — IPubSubScheduler register-and-dispatch latency across 1, 10, 100, 1000 concurrent schedules.
• SecurityBenchmarks — UadpSecurityWrapper AES-128-CTR sign+encrypt wrap/unwrap across 64, 256, 1024-byte payloads.

All suites use [MemoryDiagnoser] so every result table includes per-op allocation in the Allocated column.