Data-Driven Performance: Benchmarking in .NET with BenchmarkDotNet

2 min read 572 words

1. Introduction

When it comes to performance optimization, the first rule is: Don’t Guess. Measure.

Many developers try to measure code speed using Stopwatch, but this approach is prone to errors caused by JIT (Just-In-Time) compilation, background noise, and Garbage Collection cycles.

BenchmarkDotNet is the industry-standard library for benchmarking .NET code. It handles the complexity of “warming up” the code and providing statistically significant results.

---

2. Setting Up Your First Benchmark

BenchmarkDotNet works best in a dedicated Console Application targeting Release mode.

1. Install the NuGet Package

dotnet add package BenchmarkDotNet

2. Create the Benchmark Class

[MemoryDiagnoser] // Tracks memory allocations
public class StringBenchmark
{
    private const string Text = "Hello, World!";

    [Benchmark]
    public string UseStringConcat() => Text + " " + Text;

    [Benchmark]
    public string UseStringBuilder()
    {
        var sb = new StringBuilder();
        sb.Append(Text);
        sb.Append(" ");
        sb.Append(Text);
        return sb.ToString();
    }
}

3. Run the Benchmark

In Program.cs:

BenchmarkRunner.Run<StringBenchmark>();

CRITICAL: You must run your console app in Release mode:

dotnet run -c Release

---

3. Understanding the Results

BenchmarkDotNet generates a clean table in your console and also saves results in Markdown, CSV, and HTML formats.

Key columns to watch:

• Mean: The average time taken for one execution.
• Allocated: How much memory was allocated on the managed heap. This is often more important than speed, as high allocations trigger more frequent Garbage Collections.

---

4. Advanced Features

Parameters

You can test your code against different input sizes using the [Params] attribute:

[Params(10, 100, 1000)]
public int N { get; set; }

Multiple Runtimes

You can compare how your code performs on .NET 6 vs .NET 8 vs .NET Framework:

[SimpleJob(RuntimeMoniker.Net80)]
[SimpleJob(RuntimeMoniker.Net60)]
public class MyBenchmark { ... }

---

5. Best Practices

1. Run in Release mode: Debug mode contains extra overhead that makes measurements useless.
2. Close background apps: Slack, Chrome, and Teams can cause spikes that jitter your results.
3. Keep it isolated: A benchmark should test a small, specific unit of logic. Avoid benchmarking external database calls or network requests.
4. Use [MemoryDiagnoser]: Always track allocations. Often, a “slower” method that allocates 0 bytes is better for your app’s overall health than a “fast” one that allocates megabytes.

---

6. Conclusion

BenchmarkDotNet turns performance tuning from a “gut feeling” into a scientific process. By incorporating benchmarks into your workflow, you can confidently prove that your optimizations actually work.

Before your next refactor for performance, write a benchmark—you might be surprised by the results!