Transaction benchmark methodology
A transaction benchmark needs a declared workload, successful transaction count and persistence policy. A throughput number alone does not establish application suitability.
Report the workload
Section titled “Report the workload”Include transaction definitions and mix, scale, clients, concurrency, run duration, ramp-up and warmup. Count committed transactions separately from attempted or failed transactions.
Preserve correctness
Section titled “Preserve correctness”Check application invariants and result correctness before interpreting throughput. Record transaction isolation, retry policy and every persistence setting. A configuration that skips durable logging is not a durable-commit comparison.
Naming
Section titled “Naming”A workload inspired by TPC-C is not automatically an audited TPC-C result. Avoid implying certification or comparability unless the required specification and disclosures are satisfied.
Reproduction
Section titled “Reproduction”If the exact driver or workload is unavailable, the number cannot be independently reproduced. Do not use that result as a current homepage comparison.
This site uses Google Analytics cookies to count visits. Privacy · License