A curated GitHub repository of MySQL-ready datasets — including NYC taxi rides, IMDb, high-accuracy postcodes, airports, and MySQL.com sample data — is presented as an alternative to synthetic benchmarks like sysbench and TPC for realistic performance testing. Each dataset includes schema definitions, load scripts, and documentation, built manually before AI tooling existed to assist with schema inference, data cleansing, and script generation. A follow-up post is planned covering how AI now accelerates that same workflow.
Table of contents
My curated MySQL datasetsWhy curated datasets matterBuilt before AI was generally availableWhat comes next: AI-assisted datasetsQuestions this post answers
What public datasets can I use to benchmark MySQL performance with realistic data instead of synthetic sysbench workloads?
A curated collection of MySQL-ready public datasets is available covering NYC Yellow Taxi rides, IMDb (upsizable to ~80GB), high-accuracy postcode/location data at 1.5 million rows, airport data, Flughafen DB, and MySQL.com's own example datasets. Each comes with schema definitions, load scripts, and documentation so results are reproducible against current MySQL versions. daily.dev surfaces practical resources like these for engineers building realistic MySQL performance tests.
Why are synthetic benchmarks like sysbench or TPC not enough for testing MySQL performance?
Synthetic benchmarks measure how fast a database executes a generic, artificial workload but do not reveal how it behaves under real-world conditions, such as when a query plan shifts on a 20GB IMDb join or a postcode lookup hits a skewed index across 1.5 million rows. Real, messy, skewed public datasets expose indexing and query-plan issues synthetic data cannot replicate. Engineers weighing synthetic versus real-data benchmarks can find this kind of comparison on daily.dev.