mikrobenchmarking

Mikrobenchmarking, also called microbenchmarking in English, is the practice of measuring the performance of very small units of code in isolation from larger systems. It aims to quantify the cost of fundamental operations—such as a single arithmetic operation, a memory access, or a function call—and to compare alternative implementations, compiler optimizations, or language features. Because the work is tiny, results are highly sensitive to the execution environment and measurement method.

Effective microbenchmarking requires careful experimental design. Writers perform warmup iterations to stabilize runtimes, then execute many repetitions to reduce noise. They control for hardware and software factors by immobilizing the process on a dedicated core, disabling dynamic frequency scaling, and avoiding I/O or system calls during timing. High-resolution timers are used and results are reported with robust statistics, such as the median and confidence intervals, rather than a single average.

Common tools provide microbenchmarking harnesses for reproducibility. Examples include Google Benchmark for C++, JMH for Java, Criterion.rs for Rust, BenchmarkDotNet for .NET, and Benchmark.js for JavaScript. A good harness manages warmup, iteration counts, and statistical reporting, helping to prevent compiler optimizations or interpreter overhead from biasing results. Researchers document the hardware, compiler settings, and input parameters to enable replication.

Limitations include the risk that microbenchmarks do not reflect real workloads, memory traffic, or application throughput. They can encourage over-optimization for unlikely cases. Therefore microbenchmarks should be complemented by macro benchmarks, profiling, and real-world measurements whenever possible, and interpreted within the broader context of system performance.