◀Table of Contents

Simple Java Stream Benchmark

This application is a small benchmark of the Java Stream API. It demonstrates how the Graal compiler can achieve better performance for highly abstracted programs like those using Streams, Lambdas, or other Java features.

Preparation

1. Download GraalVM, unzip the archive, export the GraalVM Home directory as the $JAVA_HOME and add $JAVA_HOME/bin to the PATH environment variable: On Linux:

    export JAVA_HOME=/home/${current_user}/path/to/graalvm
    export PATH=$JAVA_HOME/bin:$PATH
   

   On macOS:

    export JAVA_HOME=/Users/${current_user}/path/to/graalvm/Contents/Home
    export PATH=$JAVA_HOME/bin:$PATH
   

   On Windows:

    setx /M JAVA_HOME "C:\Progra~1\Java\<graalvm>"
    setx /M PATH "C:\Progra~1\Java\<graalvm>\bin;%PATH%"
   

   Note that your paths are likely to be different depending on the download location.

2. Download or clone the repository and navigate into the java-simple-stream-benchmark directory:

    git clone https://github.com/graalvm/graalvm-demos
    cd graalvm-demos/java-simple-stream-benchmark
   

3. Build the benchmark. You can manually execute mvn package, but there is also a build.sh script included for your convenience:

    ./build.sh
   

Now you are all set to execute the benchmark and compare the results between different JVMs.

Running the Benchmark

To run the benchmark, you need to execute the target/benchmarks.jar file. You can run it with the following command:

java -jar target/benchmarks.jar

If you would like to run the benchmark on a different JVM, you can run it with whatever java you have. However, if you just want to run it on the same JVM, but without the Graal compiler, you may add the -XX:-UseJVMCICompiler option into the same command:

java -XX:-UseJVMCICompiler -jar target/benchmarks.jar

This way, the Graal compiler will not be used as the JVMCI compiler and the JVM will use its default one.

Note about Results

The benchmark mode is AverageTime in nanoseconds per operation, which means lower numbers are better. Note that the results you see can be influenced by the hardware you are running this benchmark on, the CPU load, and other factors. Interpret them responsibly.