Skip to content

Benchmarks.md

Latest commit

 

History

History
145 lines (103 loc) · 5.38 KB

Benchmarks.md

File metadata and controls

145 lines (103 loc) · 5.38 KB

DataFlow.NET Performance Benchmarks

Generated: December 30, 2025

Executive Summary

DataFlow.NET achieves performance parity with industry standards (CsvHelper, System.Text.Json) while providing a unified API for all formats.

  • CSV: Parity with CsvHelper at 1GB scale.
  • JSON: ~40% overhead vs System.Text.Json (trade-off for unified API).
  • YAML: 38% FASTER than YamlDotNet at 1GB scale.
  • Memory: Constant low memory usage (~3-6 MB) regardless of file size (tested up to 1GB).

Environment

BenchmarkDotNet v0.15.2, Windows 11 (10.0.22631.6060/23H2)
12th Gen Intel Core i5-12450H 2.00GHz, 1 CPU, 12 logical / 8 physical cores
.NET SDK 9.0.308
Host: .NET 8.0.22 (8.0.2225.52707), X64 RyuJIT AVX2
Job: ShortRun (3 iterations, 3 warmup)

1GB Scale Competitor Comparisons

We benchmarked DataFlow.NET against industry-standard libraries using ~1GB datasets to demonstrate real-world scalability.

CSV (~1GB, 13M rows)

Library Mean Time Speed Allocated vs DataFlow Sync
DataFlow.NET (Sync) 21.8s ~46 MB/s 10.85 GB Baseline
CsvHelper 21.8s ~46 MB/s 4.97 GB Equal (PARITY!)
DataFlow.NET (Async) 28.2s ~35 MB/s 10.85 GB 1.3x slower

Insight: Use Read.CsvSync for maximum throughput on large files.

JSON (~1GB, 8M objects)

Library Method Mean Time Allocated vs DataFlow
System.Text.Json DeserializeAsyncEnumerable 9.9s 1.11 GB 1.66x faster
System.Text.Json Deserialize (Sync) 12.3s 1.28 GB 1.34x faster
DataFlow.NET Read.Json (Async) 16.5s 1.10 GB Baseline
DataFlow.NET Read.JsonSync 16.6s 1.10 GB 1.01x slower

Insight: ~40% overhead is the cost of the unified SUPRA abstraction.

YAML (~1GB, 8M documents)

Library Method Mean Time Allocated vs DataFlow Sync
DataFlow.NET (Sync) Read.YamlSync 1.92 min 53.1 GB Baseline
DataFlow.NET (Async) Read.Yaml (Async) 2.76 min 53.1 GB 1.44x slower
YamlDotNet Direct 3.11 min 53.1 GB 1.62x slower

Insight: DataFlow's streaming wrapper is significantly optimized for large multi-document files compared to raw YamlDotNet usage.

Memory Efficiency (1GB Streaming)

DataFlow.NET uses lazy streaming to keep memory usage constant (~3-6 MB) regardless of input size.

CSV Memory Profile (1GB)

%%{init: {'theme': 'base', 'themeVariables': {'xyChart': {'plotColorPalette': '#22c55e'}}}}%%
xychart-beta
    title "Streaming Memory (Smoothed) (1081 MB CSV)"
    x-axis "Data Processed (MB)" [16, 119, 222, 319, 416, 513, 611, 708, 800, 897, 994, 1081]
    y-axis "Live Memory (MB)" 0 --> 10
    line [4.2, 3.6, 3.6, 3.0, 3.0, 3.1, 4.2, 4.2, 3.5, 4.3, 4.3, 0.2]
Loading

JSON Memory Profile (1GB)

Matches System.Text.Json efficiency showing true lazy evaluation.

%%{init: {'theme': 'base', 'themeVariables': {'xyChart': {'plotColorPalette': '#22c55e, #3b82f6'}}}}%%
xychart-beta
    title "Live Memory During JSON Streaming (1063 MB File)"
    x-axis "Data Processed (MB)" [0, 106, 212, 318, 425, 531, 637, 744, 850, 956, 1063]
    y-axis "Live Memory (MB)" 0 --> 10
    line [0.0, 0.8, 5.8, 6.1, 5.4, 6.6, 6.1, 6.5, 6.6, 6.0, 0.2]
    line [0.0, 0.5, 0.6, 0.7, 0.6, 0.7, 0.9, 1.1, 1.0, 0.1, 0.0]
Loading

Legend: 🟢 DataFlow.NET | 🔵 System.Text.Json

YAML Memory Profile (1GB)

%%{init: {'theme': 'base', 'themeVariables': {'xyChart': {'plotColorPalette': '#22c55e'}}}}%%
xychart-beta
    title "Live Memory During YAML Streaming (1039 MB File)"
    x-axis "Data Processed (MB)" [0, 103, 207, 311, 415, 519, 623, 727, 831, 935, 1039]
    y-axis "Live Memory (MB)" 0 --> 10
    line [0.0, 4.8, 2.1, 6.5, 3.3, 1.0, 1.5, 0.7, 2.2, 2.7, 0.1]
Loading

Memory Stability Check

To ensure no memory leaks exist (e.g., unbounded static cache growth), we ran a leak check (10 iterations):

Format Run 1 (Delta) Run 10 (Delta) Status
JSON 0.21 MB 0.23 MB ✅ Stable (Static Overhead)
YAML 0.13 MB 0.18 MB ✅ Stable (Static Overhead)

Internal Micro-benchmarks

ObjectMaterializer

Compares DataFlow.NET's ObjectMaterializer against raw reflection.

Method Mean Allocated vs Baseline
Baseline (direct new()) 13.78 ns 48 B 1.00x
CtorSession.Create() 30.28 ns 48 B 2.20x
Reflection.SetProperties 121.71 ns 104 B 8.85x

Insight: Session-based creation is 4x faster than reflection and has the same memory footprint as direct instantiation.

Running the Benchmarks

cd benchmarks

# Run Performance Benchmarks (will generate large files)
dotnet run -c Release --project DataFlow.Data.Benchmarks -- --filter "*" --job short

# Run Memory Profiles
dotnet run -c Release --project DataFlow.Data.Benchmarks -- --profile           # CSV
dotnet run -c Release --project DataFlow.Data.Benchmarks -- --json-memory       # JSON
dotnet run -c Release --project DataFlow.Data.Benchmarks -- --yaml-memory       # YAML
dotnet run -c Release --project DataFlow.Data.Benchmarks -- --leak-check        # Stability