Some benchmarks for the stream-lib

This small project is intended to help stream-lib contributors to check the performance impact of their contribution and to discover area of improvement for the project.

Micro benchmarks must be used to compare two implementations or to track performance regressions. Do not use them to predict application behavior or to known if a given operation is fast enough. They are not realistic use cases, and they are written to be as fast as possible (no IO bottleneck, everyhing in CPU cache etc.).

Macro benchmarks try to simulate real world use cases. They can be used to get back-of-the-envelope performance estimations for real world application, or to discover if it worth optimizing something further. For example, the merge operation will most likely require reading bytes from the disk or network. If the code is fast enough to be IO bound, optimizing it is a waste of time.

Be sure to understand the meaning of a given test and how compare two numbers before trying to use it.

This project is very young, always check the code and the output before phrasing any statement. Feel free to provide feedback, improve the test suite or add new tests.

Howto

Install dependencies

Java

JDK 7 or higher is required.

jmh

The benchmarks use jmh. Unfortunately the OpenJDK team has not yet released a stable version of jmh. You have to build it by yourself and publish the artifact in your local Maven repository.

$ hg clone http://hg.openjdk.java.net/code-tools/jmh/ jmh
$ cd jmh/
$ mvn clean install -DskipTests=true

Build

$ gradle compile

Run

# List available benchmarks
$ gradle run -Pargs="-l"
# Run the whole test suite
$ gradle run