SBT plugin for running OpenJDK JMH benchmarks.
JMH is a Java harness for building, running, and analysing nano/micro/milli/macro benchmarks written in Java and other languages targeting the JVM.
Please read nanotrusting nanotime and other blog posts on micro-benchmarking (or why most benchmarks are wrong) and make sure your benchmark is valid, before you set out to implement your benchmarks.
The latest published plugin version is: 
| Plugin version | Shipped JMH version |
|---|---|
0.2.7 (auto plugin) |
1.12 |
0.2.6 (auto plugin) |
1.11.3 |
0.2.5 (auto plugin) |
1.11 |
0.2.4 (auto plugin) |
1.10.3 |
0.2.1 (auto plugin) |
1.10 |
0.2.0 (auto plugin) |
1.9.1 |
0.1.15 (auto plugin) |
1.9.1 |
0.1.14 |
1.8.0 |
| ... | ... |
Not interesting versions are skipped in the above listing. Always use the newest which has the JMH version you need. You should stick to the latest version at all times anyway of course.
Just use the Typesafe Activator to get the template downloaded:
activator new PROJECT_NAME sbt-jmh-seed
And start writing benchmarks!
Hint: You have to trigger jmh:compile, so that jmh creates the BenchmarkList.
Since sbt-jmh is an AutoPlugin all you need to do in order to activate it in
your project is to add the below line to your project/plugins.sbt file:
// project/plugins.sbt
addSbtPlugin("pl.project13.scala" % "sbt-jmh" % "0.2.7")and enable it in the projects where you want to (useful in multi-project builds, as you can enable it only where you need it):
// build.sbt
enablePlugins(JmhPlugin)If you define your project in a Build.scala, you also need the following import:
import pl.project13.scala.sbt.JmhPluginYou can read more about auto plugins in sbt on it's documentation page.
Write your benchmarks in src/main/scala. They will be picked up and instrumented by the plugin.
JMH has a very specific way of working (it generates loads of code), so you should prepare a separate project for your benchmarks. In it, just type run in order to run your benchmarks.
All JMH options work as expected. For help type run -h. Another example of running it is:
jmh:run -i 3 -wi 3 -f1 -t1 .*FalseSharing.*
Which means "3 iterations" "3 warmup iterations" "1 fork" "1 thread". Please note that benchmarks should be usually executed at least in 10 iterations (as a rule of thumb), but more is better.
For "real" results we recommend to at least warm up 10 to 20 iterations, and then measure 10 to 20 iterations again. Forking the JVM is required to avoid falling into specific optimisations (no JVM optimisation is really "completely" predictable)
Please invoke run -h to get a full list of run as well as output format options.
Useful hint: If you plan to aggregate the collected data you should have a look at the available output formats (-lrf).
For example it's possible to keep the benchmark's results as csv or json files for later regression analysis.
The examples are scala-fied examples from the original JMH repo, check them out, and run them!
The results will look somewhat like this:
...
[info] # Run progress: 92.86% complete, ETA 00:00:15
[info] # VM invoker: /Library/Java/JavaVirtualMachines/jdk1.7.0_60.jdk/Contents/Home/jre/bin/java
[info] # VM options: <none>
[info] # Fork: 1 of 1
[info] # Warmup: 2 iterations, single-shot each
[info] # Measurement: 3 iterations, single-shot each
[info] # Threads: 1 thread, will synchronize iterations
[info] # Benchmark mode: Single shot invocation time
[info] # Benchmark: org.openjdk.jmh.samples.JMHSample_02_BenchmarkModes.measureSingleShot
[info] # Warmup Iteration 1: 100322.000 us
[info] # Warmup Iteration 2: 100556.000 us
[info] Iteration 1: 100162.000 us
[info] Iteration 2: 100468.000 us
[info] Iteration 3: 100706.000 us
[info]
[info] Result : 100445.333 ±(99.9%) 4975.198 us
[info] Statistics: (min, avg, max) = (100162.000, 100445.333, 100706.000), stdev = 272.707
[info] Confidence interval (99.9%): [95470.135, 105420.532]
[info]
[info]
[info] # Run progress: 96.43% complete, ETA 00:00:07
[info] # VM invoker: /Library/Java/JavaVirtualMachines/jdk1.7.0_60.jdk/Contents/Home/jre/bin/java
[info] # VM options: <none>
[info] # Fork: 1 of 1
[info] # Warmup: 2 iterations, single-shot each, 5000 calls per batch
[info] # Measurement: 3 iterations, single-shot each, 5000 calls per batch
[info] # Threads: 1 thread, will synchronize iterations
[info] # Benchmark mode: Single shot invocation time
[info] # Benchmark: org.openjdk.jmh.samples.JMHSample_26_BatchSize.measureRight
[info] # Warmup Iteration 1: 15.344 ms
[info] # Warmup Iteration 2: 13.499 ms
[info] Iteration 1: 2.305 ms
[info] Iteration 2: 0.716 ms
[info] Iteration 3: 0.473 ms
[info]
[info] Result : 1.165 ±(99.9%) 18.153 ms
[info] Statistics: (min, avg, max) = (0.473, 1.165, 2.305), stdev = 0.995
[info] Confidence interval (99.9%): [-16.988, 19.317]
[info]
[info]
[info] Benchmark Mode Samples Mean Mean error Units
[info] o.o.j.s.JMHSample_22_FalseSharing.baseline thrpt 3 692.034 179.561 ops/us
[info] o.o.j.s.JMHSample_22_FalseSharing.baseline:reader thrpt 3 199.185 185.188 ops/us
[info] o.o.j.s.JMHSample_22_FalseSharing.baseline:writer thrpt 3 492.850 7.307 ops/us
[info] o.o.j.s.JMHSample_22_FalseSharing.contended thrpt 3 706.532 293.880 ops/us
[info] o.o.j.s.JMHSample_22_FalseSharing.contended:reader thrpt 3 210.202 277.801 ops/us
[info] o.o.j.s.JMHSample_22_FalseSharing.contended:writer thrpt 3 496.330 78.508 ops/us
[info] o.o.j.s.JMHSample_22_FalseSharing.hierarchy thrpt 3 1751.941 222.535 ops/us
[info] o.o.j.s.JMHSample_22_FalseSharing.hierarchy:reader thrpt 3 1289.003 277.126 ops/us
[info] o.o.j.s.JMHSample_22_FalseSharing.hierarchy:writer thrpt 3 462.938 55.329 ops/us
[info] o.o.j.s.JMHSample_22_FalseSharing.padded thrpt 3 1745.650 83.783 ops/us
[info] o.o.j.s.JMHSample_22_FalseSharing.padded:reader thrpt 3 1281.877 47.922 ops/us
[info] o.o.j.s.JMHSample_22_FalseSharing.padded:writer thrpt 3 463.773 104.223 ops/us
[info] o.o.j.s.JMHSample_22_FalseSharing.sparse thrpt 3 1362.515 461.782 ops/us
[info] o.o.j.s.JMHSample_22_FalseSharing.sparse:reader thrpt 3 898.282 415.388 ops/us
[info] o.o.j.s.JMHSample_22_FalseSharing.sparse:writer thrpt 3 464.233 49.958 ops/us
It is possible to hand over the running of JMH to an App implemented by you, which allows you to programmatically
access all test results and modify JMH arguments before you actually invoke it.
To use a custom runner class with runMain, simply use it: jmh:runMain com.example.MyRunner -i 10 .* –
an example for this is available in src/sbt-test/sbt-jmh/runMain (open the test file).
To replace the runner class which is used when you type jmh:run, you can set the class in your build file –
an example for this is available in src/sbt-test/sbt-jmh/custom-runner (open the build.sbt file).
This plugin is released under the Apache 2.0 License
Yes, pull requests and opening issues is very welcome!
Please test your changes using sbt scripted.