SQL Arena

SQL Arena is an independent project that aims to provide comparative data between database vendors. This should help you make informed decisions about which database to use for your next project and tell you where to expect issues.

Data rendered on this website is collected using a tool called dbprove.

All tooling used by SQL Arena are open source under the Apache License. People interested in databases are invited to contribute (see: DBProve).

In the cases where the DeWitt Clause allows it — performance metrics will be measured. For other cases, vendors will either be excluded or measurements not related to performance will be published.

Data is divided into components that roughly match the typical, architectural stack of a relational database.

Queries are bundled into workloads (for example: TPC-H). After each query run a post-hoc analysis is done to determine how well the query was executed by the database engine. The focus is not just on runtime, but also on what we can say about the database engine based on the observed behaviour of the query

The automated Analysis Includes such things as:

• Runtimes broken into their components (when the engine supports it)
• Query Plan shapes and relational operator usage
• Latency Measurements
• CPU cycle breakdown when viable — falling back estimation when the engine is lacking instrumentation.
• I/O detailed analysis using whatever instrumentation is available

We are glad you asked!

You can contribute results to this site via our result repository.

If you feel like hacking code - you can even file Pull Requests against DBProve

More details are here: Contributing

This site is meant to be educational. It is intended as a way to compare multiple databases against each other in as neutral and objective a way as possible.

We are always looking to improve the site and our measurement. Comments, suggestions and improvements are most welcome and can be directed to Thomas Kejser at my LinkedIn account or via email.

All data rendered on this site is made with DBProve or gathered from the analysis of contributors.