This example demonstrates how to write simple micro-benchmarks, based on JMH, for JAX-RS application. This particular example shows how to obtain throughput of a simple ‘Hello World’ application.
The application consists of one resource, HelloWorldResource with three resource methods (one method per HTTP GET, POST and PUT methods). To make it more interesting the resource also acts like sub-resource locator which allows us to measure the difference between processing of normal resource methods and sub-resource locators.
This application is solely focused on showing how to write benchmarks and it doesn't start any server that would allow one to send request to the HelloWorldResource.
Main class that illustrates writing benchmarks for JAX-RS applications is HelloWorldBenchmark. Measurements are not based on any container such as Servlet or Grizzly container. Requests are directly invoked on ApplicationHandler class from Jersey Server module to eliminate any network influences. Still, all JAX-RS features, like filters and interceptors, are taken into consideration when the benchmark is executed.
It's also possible to run the benchmark on older versions of Jersey, e.g. 2.15, which had known performance issues with sub-resource locators. The recommended approach is to run the benchmarks without any changes at first and then modify the pom.xml to switch version of Jersey to 2.15 and run the benchmarks again to see the (big) difference. Look for jersey-bom artifact in pom.xml to see how the switch should be done.
The execution takes around 2 minutes and you can see partial results (for each combination of parameters) of every iteration. At the end JMH gives you whole summary. For example, Jersey 2.16 gives the following results:
# Run complete. Total time: 00:01:41 Benchmark (method) (path) Mode Cnt Score Error Units HelloWorldBenchmark.measure GET helloworld thrpt 8 74343.864 ± 4814.979 ops/s HelloWorldBenchmark.measure GET helloworld/locator thrpt 8 54137.102 ± 8996.766 ops/s HelloWorldBenchmark.measure POST helloworld thrpt 8 45173.853 ± 5349.363 ops/s HelloWorldBenchmark.measure POST helloworld/locator thrpt 8 37144.797 ± 4699.782 ops/s HelloWorldBenchmark.measure PUT helloworld thrpt 8 45945.974 ± 4116.752 ops/s HelloWorldBenchmark.measure PUT helloworld/locator thrpt 8 36345.667 ± 5929.480 ops/s
There are two ways how to run the micro-benchmark. Either via Maven exec:exec
mvn clean install exec:exec
or you can build the benchmark JAR at first
mvn clean install
and then invoking it using java command
java -jar target/benchmark.jar
JMH has excellent overview of it's features presented as list of different samples. You can find them on JMH Samples. The general overview of JMH is available at their home page.