Skip to end of metadata
Go to start of metadata

Whirr provides a way to run benchmarks on a Hadoop cluster. (This was introduced in WHIRR-92.) This allows you to see how a particular cluster performs on a particular provider, and can provide a basis for tuning.

The following command will start a Hadoop cluster then run a suite of benchmarks (currently Terasort as DFSIO) against it, before tearing it down:

mvn verify -Pintegration \
  -DargLine="-Dwhirr.test.provider=<cloud-provider> -Dwhirr.test.identity=<cloud-provider-user> -Dwhirr.test.credential=<cloud-provider-secret-key>" \
  -Dit.test=HadoopBenchmarkSuite

If you want to provide extra properties, you can do so as follows by specifying the config option. Here we use a file called .whirr-test.properties in our home directory:

mvn verify -Pintegration \
  -DargLine="-Dwhirr.test.provider=<cloud-provider> -Dwhirr.test.identity=<cloud-provider-user> -Dwhirr.test.credential=<cloud-provider-secret-key> -Dconfig=.whirr-test.properties" \
  -Dit.test=HadoopBenchmarkSuite

You can run a single benchmark as follows (note that this launches and tears down a cluster for this one test):

mvn verify -Pintegration \
  -DargLine="-Dwhirr.test.provider=<cloud-provider> -Dwhirr.test.identity=<cloud-provider-user> -Dwhirr.test.credential=<cloud-provider-secret-key>" \
  -Dit.test=HadoopServiceTestDFSIOBenchmark

Some tests take properties to control their behavior. E.g. you can specify the amount of data to sort in Terasort by specifying terasortBytesPerNode:

mvn verify -Pintegration \
  -DargLine="-Dwhirr.test.provider=<cloud-provider> -Dwhirr.test.identity=<cloud-provider-user> -Dwhirr.test.credential=<cloud-provider-secret-key> -DterasortBytesPerNode=100000000" \
  -Dit.test=HadoopBenchmarkSuite
  • No labels