Benchmarks
All benchmarks run on .NET 10 using BenchmarkDotNet. Tested against Mediator 3.0.1 (source-generator based), DispatchR 2.1.1, and MediatR 14.1.
Latency
| Operation | DSoft | Mediator (SG) | DispatchR | MediatR |
|---|---|---|---|---|
Send() |
7.2 ns | 12.2 ns | 33.4 ns | 41.3 ns |
Send() (5 behaviors) |
15.6 ns | 36.8 ns | 54.1 ns | 153.1 ns |
Publish() |
4.5 ns | 10.6 ns | 35.7 ns | 123.4 ns |
CreateStream() |
45.5 ns | 44.7 ns | 68.1 ns | 122.9 ns |
| Cold Start | 1.62 µs | 9.91 µs | 1.88 µs | 3.24 µs |
Key Takeaways
- Send is ~1.7× faster than the next fastest (Mediator SG) and ~5.7× faster than MediatR
- Send with 5 behaviors shows the pipeline chain stays efficient under load — only ~1.7 ns per behavior hop (8.4 ns total overhead for 5 behaviors)
- Publish is ~0.7 ns overhead — effectively a direct call to the handler array
- Cold Start at 1.62 µs means precompiled pipelines warm up in under 2 microseconds
Allocations
| Operation | DSoft | Mediator (SG) | DispatchR | MediatR |
|---|---|---|---|---|
Send() |
72 B | 72 B | 72 B | 272 B |
Send() (5 behaviors) |
72 B | 72 B | 72 B | 1,088 B |
Publish() |
0 B | 0 B | 0 B | 768 B |
CreateStream() |
232 B | 232 B | 232 B | 624 B |
Key Takeaways
- 72 B per Send is the
ValueTask<T>boxing cost — the pipeline itself allocates nothing - Adding 5 behaviors does not increase allocation — interface dispatch chains are zero-alloc
- Publish is 0 B — notification dispatch is fully allocation-free
- MediatR allocates 1,088 B with 5 behaviors due to delegate chains (
RequestHandlerDelegate<T>)
Pipeline Allocation Breakdown
| Component | DSoft | MediatR |
|---|---|---|
| Handler resolution | 0 B (singleton) | ~40 B (transient GetService) |
| Pipeline chain | 0 B (interface dispatch) | ~160 B per behavior (delegate) |
| Return value | 72 B (ValueTask box) |
72 B (Task alloc) |
| 5-behavior total | 72 B | 1,088 B |
Feature Comparison
| Feature | DSoft | Mediator (SG) | DispatchR | MediatR |
|---|---|---|---|---|
| Source generators | ✔️ | ✔️ | ❌ | ❌ |
| Native AOT compatible | ✔️ | ✔️ | ❌ | ❌ |
| Reflection-free hot path | ✔️ | ✔️ | ❌ | ❌ |
| Zero-alloc pipeline | ✔️ | ✔️ | ✔️ | ❌ |
| Auto-Singleton handlers | ✔️ | ❌ | ❌ | ❌ |
| Self-handling requests | ✔️ | ❌ | ❌ | ❌ |
Runtime-typed Send(object) |
✔️ | ❌ | ❌ | ✔️ |
| Compile-time pipeline | ✔️ | ✔️ | ❌ | ❌ |
| MediatR-style API | ✔️ | ✔️ | ❌ | ✔️ |
Realistic Pipeline (Enterprise)
Measures all 4 libraries through a production-representative pipeline: Validation → Logging → Metrics → async database write with 3 pipeline behaviors, full DI, and simulated async I/O. Each library has its own DirectCall_WithPipeline fair baseline measured in the same isolated BenchmarkDotNet process.
| Benchmark | Latency | Allocated | vs DirectCall |
|---|---|---|---|
| DSoft — Realistic Pipeline | 666.9 ns | 255 B | 0.99× |
| DispatchR — Realistic Pipeline | 666.7 ns | 255 B | 1.01× |
| Mediator SG — Realistic Pipeline | 718.0 ns | 397 B | 1.06× |
| MediatR — Realistic Pipeline | 857.1 ns | 1,032 B | 1.20× |
Reproducing
Full BenchmarkDotNet results and source code are available in the /benchmarks folder.
dotnet run -c Release --project benchmarks/DSoftStudio.Mediator.Benchmarks
Individual library runs:
benchmarks/run-dsoft.cmd
benchmarks/run-mediatr.cmd
benchmarks/run-dispatchr.cmd
benchmarks/run-mediatorsg.cmd
See Also
- Performance Design — explains the techniques behind these numbers
- Migration from MediatR — step-by-step guide to switch
- Pipeline Behaviors — zero-allocation behavior chains explained
- ADR-0001: Architecture Overview — full architecture decision record