Beyond benchmarks: systematically tuning Go with the Taguchi Method

In this talk, we introduce a Go framework implementing the Taguchi method, enabling structured experimentation in systems engineering workflows. The framework uses orthogonal arrays to efficiently sample parameter spaces, reducing the number of experiments while preserving the ability to extract meaningful structure from results.

Through a concrete example: tuning a parallel workload by varying goroutine count, workload characteristics, and execution configuration - we explore how parameter interactions surface in practice, including trade-offs between concurrency, scheduling, and infrastructure such as multi-core scaling versus isolated-core execution.

This talk shows how Taguchi experimental design can be used as a practical tool for reasoning about system behavior in settings where interactions between parameters make intuition or single-variable tuning insufficient.