Peer benchmarks

This page compares the same two-rule model across Jeninnet.Validation, FluentValidation 12.2, and DataAnnotations. It is a focused microbenchmark, not a full application benchmark.

Model and rules

The model has one string and one integer member:

public sealed class PeerModel
{
    [Required]
    public string? Name { get; set; }

    [Range(18, 120)]
    public int Age { get; set; }
}

All three paths apply the same rules: Name is required and at most 10 characters; Age is between 18 and 120. The Jeninnet and FluentValidation validators use their fluent APIs; DataAnnotations uses Validator.TryValidateObject(..., validateAllProperties: true).

Results

ShortRun job on the repository’s reference machine (Windows 11, Intel Core i7-8850H, .NET 10.0.14, x64 RyuJIT). Lower is better; allocations are managed bytes per operation. This is a historical measurement; current builds use .NET 11 SDK 11.0.100 while the benchmark project targets the net10.0 baseline, so re-run before quoting it as a current performance claim.

Method Mean Allocated
Jeninnet — valid 81.7 ns 0 B
FluentValidation — valid 226.7 ns 632 B
DataAnnotations — valid 848.3 ns 1,408 B
Jeninnet — invalid 1,048.5 ns 1,976 B
FluentValidation — invalid 2,456.4 ns 4,880 B
DataAnnotations — invalid 1,488.9 ns 1,752 B

On the valid path, Jeninnet is about 2.8× faster than FluentValidation and allocates nothing. The invalid path is dominated by error collection and message formatting; it is still about 2.3× faster than FluentValidation. DataAnnotations allocates slightly less than Jeninnet on this invalid path because it stores ValidationResult objects without Jeninnet’s parameter dictionary; that is a payload difference, not a throughput win.

Reproduce

dotnet run -c Release --project src/Jeninnet.Validation.Benchmarks \
  -- --filter "*PeerComparisonBenchmarks*"

The benchmark class is PeerComparisonBenchmarks in src/Jeninnet.Validation.Benchmarks/PeerComparisonBenchmarks.cs.

Caveats

  • ShortRunJob is intentionally quick and noisy; use a full job for release claims.
  • Results are machine-, runtime-, and model-specific. Re-run on your hardware.
  • The benchmark isolates validation execution. It excludes model binding, HTTP, database calls, localization lookup, and logging.
  • DataAnnotations defaults may short-circuit differently from a custom ValidationContext; this benchmark keeps the default TryValidateObject behavior.