SQLBench Runners

This repository contains code to run the SQLBench-H and SQLBench-DS benchmarks using a variety of query engines.

All SQLBench benchmarks following the same format:

• There is one directory containing all data files or directories to be registered as SQL tables
• There is one directory containing all SQL queries to be executed against these tables. The queries are numbered sequentially, starting with query 1. File names are q1.sql, q2.sql, and so on. Each file may contain multiple SQL statements.

Creating Docker Images

There is one script per query engine for building the Docker image. For example, the following commands can be used to build a Docker image for DuckDB.

cd duckdb
./build-docker.sh

Running Single Node Benchmarks

All benchmarks are run in a constrained containerized environment for fairness. The environment is constrained by specifying --cpus and -m.

Here is an example of running the DataFusion Python benchmarks. See the README for each engine for specific instructions.

docker run \
  --cpus 16 \
  -m 64GB \
  -v /mnt/bigdata/tpch/sf10-parquet/:/data \
  -v `pwd`/../sqlbench-h/queries/sf\=10/:/queries \
  -it sqlbench/datafusion-python /data /queries 22

Running Distributed Benchmarks

Coming soon. The basic idea is to use Kubernetes.

Legal Stuff

SQLBench is a Non-TPC Benchmark. Any comparison between official TPC Results with non-TPC workloads is prohibited by the TPC.

TPC, TPC Benchmark, TPC-H, and TPC-DS are trademarks of the Transaction Processing Performance Council.