Skip to content

Latest commit

 

History

History
58 lines (35 loc) · 3.14 KB

build-hibench.md

File metadata and controls

58 lines (35 loc) · 3.14 KB
Raw

Build All

To simply build all modules in HiBench, use the below command. This could be time consuming because the hadoopbench relies on 3rd party tools like Mahout and Nutch. The build process automatically downloads these tools for you. If you won't run these workloads, you can only build a specific framework to speed up the build process.

mvn -Dspark=2.1 -Dscala=2.11 clean package

Build a specific framework benchmark

HiBench 6.0 supports building only benchmarks for a specific framework. For example, to build the Hadoop benchmarks only, we can use the below command:

mvn -Phadoopbench -Dspark=2.1 -Dscala=2.11 clean package

To build Hadoop and spark benchmarks

mvn -Phadoopbench -Psparkbench -Dspark=2.1 -Dscala=2.11 clean package

Supported frameworks includs: hadoopbench, sparkbench, flinkbench, stormbench, gearpumpbench.

Specify Scala Version

To specify the Scala version, use -Dscala=xxx(2.10 or 2.11). By default, it builds for scala 2.11.

mvn -Dscala=2.10 clean package

tips: Because some Maven plugins cannot support Scala version perfectly, there are some exceptions.

  1. No matter what Scala version is specified, the module (gearpumpbench/streaming) is always built in Scala 2.11.
  2. When the spark version is specified to 2.0 or higher, the module (sparkbench/streaming) is only supported for Scala 2.11.

Specify Spark Version

To specify the spark version, use -Dspark=xxx(1.6, 2.0, 2.1, 2.2, 2.3 or 2.4). By default, it builds for spark 2.0

mvn -Psparkbench -Dspark=1.6 -Dscala=2.11 clean package

tips: when the spark version is specified to spark2.0(1.6) , the scala version will be specified to scala2.11(2.10) by default . For example , if we want use spark2.0 and scala2.11 to build hibench. we just use the command mvn -Dspark=2.0 clean package , but for spark2.0 and scala2.10 , we need use the command mvn -Dspark=2.0 -Dscala=2.10 clean package . Similarly , the spark1.6 is associated with the scala2.10 by default.

Build a single module

If you are only interested in a single workload in HiBench. You can build a single module. For example, the below command only builds the SQL workloads for Spark.

mvn -Psparkbench -Dmodules -Psql -Dspark=2.1 -Dscala=2.11 clean package

Supported modules includes: micro, ml(machine learning), sql, websearch, graph, streaming, structuredStreaming(spark 2.0 or higher) and dal.

Build Structured Streaming

For Spark 2.0 or higher versions, we add the benchmark support for Structured Streaming. This is a new module which cannot be compiled in Spark 1.6. And it won't get compiled by default even if you specify the spark version as 2.0 or higher. You must explicitly specify it like this:

mvn -Psparkbench -Dmodules -PstructuredStreaming clean package

Build DAL on Spark

By default the dal module will not be built and needs to be enabled explicitly by adding "-Dmodules -Pdal", for example:

mvn -Psparkbench -Dmodules -Pml -Pdal -Dspark=2.0 -Dscala=2.11 clean package

Currently there is only one workload KMeans available in DAL. To run the workload, install DAL and setup the environment by following https://github.com/intel/daal