Skill: benchmarking

What I do

I provide Go-specific benchmarking expertise to measure and optimise code performance. I focus on writing reliable benchmarks using the testing package and analysing results to identify bottlenecks.

When to use me

• When comparing the performance of multiple implementations
• When verifying the impact of an optimisation
• When identifying hotspots in performance-critical code paths

Core principles

1. Isolation: Run benchmarks in a stable environment to minimise noise.
2. Reliability: Use b.ResetTimer() to exclude setup overhead and b.ReportAllocs() to track memory allocations.
3. Statistical significance: Use tools like benchstat to compare results across multiple runs.
4. Realistic data: Use representative input sizes to avoid misleading results from small or trivial datasets.

Patterns & examples

Standard benchmark function:

func BenchmarkProcessData(b *testing.B) {
    data := setupTestData()
    b.ResetTimer()
    b.ReportAllocs()

    for i := 0; i < b.N; i++ {
        ProcessData(data)
    }
}

Table-driven benchmark:

func BenchmarkAlgorithm(b *testing.B) {
    benchmarks := []struct {
        name string
        size int
    }{
        {"Small", 10},
        {"Medium", 100},
        {"Large", 1000},
    }
    for _, bm := range benchmarks {
        b.Run(bm.name, func(b *testing.B) {
            data := generateData(bm.size)
            b.ResetTimer()
            for i := 0; i < b.N; i++ {
                Algorithm(data)
            }
        })
    }
}

Comparing results: Use go test -bench . -count 5 > old.txt and go test -bench . -count 5 > new.txt, then run benchstat old.txt new.txt to see the percentage change.

Anti-patterns to avoid

• ❌ Looping manually: Always use b.N for the loop count. Hardcoding iterations leads to unreliable timing.
• ❌ Compiler optimisations: Ensure the result of the function under test is used (e.g., assigned to a package-level variable) to prevent the compiler from eliding the call.
• ❌ Ignoring allocations: High memory allocation counts often indicate performance issues that timing alone might miss.

KB Reference

~/vaults/baphled/3. Resources/Knowledge Base/AI Development System/Skills/Performance-Profiling/Benchmarking.md

Related skills

• profiling: For deep dives into where time or memory is spent
• performance: General optimisation principles and techniques
• golang: For idiomatic Go patterns and standard library usage