Alluxio Microbenchmarks

This module contains microbenchmarks written in JMH for various Alluxio components.

Building the benchmarks

This module is built along with others in a normal full project build. Alternatively, you can build this module alone by running the following build command from the repository root:

$ mvn -DskipTests -Dcheckstyle.skip -Dlicense.skip -Dfindbugs.skip package -pl microbench

This will build an uber jar named benchmarks.jar under microbench/target containing all benchmarks in this module.

Running the benchmarks

The generated benchmarks.jar is an executable jar, so you can use the following command to run the benchmarks:

$ java -jar microbench/target/benchmarks.jar <benchmark name>

Replace <benchmark name> with a benchmark name, or a regex pattern to include a set of benchmarks whose name matches the pattern.

Useful options

JMH support dozens of options that let you fine tune the execution of the benchmarks and the generation of the report. Use option -h to get usage and a list of options supported by JMH.

Selecting benchmarks

Use option -l [pattern] to see a list of all available benchmarks. You can further specify a regex pattern to list benchmarks matching the pattern:

$ java -jar microbench/target/benchmarks.jar -l rpc

This lists all benchmarks whose fully qualified class name contains the word rpc.

Use -e pattern to exclude benchmarks from running.

Controlling execution of benchmarks

There are plenty of options that can be used to control how a benchmark should run. Only the most often used ones are listed here.

Some benchmarks are parameterized. Use option -lp to get available parameters of the benchmarks. Use -p param=value to specify a particular parameter.

Use -i and -wi to specify how many measurement and warmup iterations to run, respectively. Use -r and -w to specify how long each measurement and warmup iteration should run for, respectively.

Use -f to specify how many times to fork each benchmark. Forks are repeated runs of both the warmup phase and the measurement phase under the same set of parameters. Make sure to at least fork once to get more accurate results, unless you are debugging the benchmark from an IDE.

Result and output

Use -rff to specify an output file to which the results will be saved for later data processing. Combine with -rf to specify the output format (CSV or JSON, etc).

Use -v to enable verbose output.