Benchmark Methodology

A useful serialization benchmark records enough context to reproduce the workload and avoid comparing different semantics as if they were equivalent.

Required context

• repository commit and dependency versions;
• hardware, operating system, runtime version, and relevant environment variables;
• schema and object population;
• xlang or native mode, compatibility mode, reference tracking, and registration;
• warm-up, forks/processes, iterations, duration, and concurrency;
• serialized representation and operation measured;
• exact command and raw result location.

Interpret the reports

Compare one operation and equivalent data semantics. Do not compare an xlang payload with a native payload or a full object reconstruction with Row Format field access without naming that semantic difference. JIT- and codegen-based runtimes require representative warm-up.

Run the active benchmark harness for decisions that depend on current code. Checked-in reports are evidence for the commit and environment they record.