Makes interoperability between H2O and Spark trivial.

• Spark 1.0.0 (SQL component required)
• Tachyon 0.4.1
• Java 1.6+

• First compile latest version of spark with SQL component

git clone spark
cd spark
sbt/sbt assembly publish-local

• For Tachyon support please download Tachyon 0.4.1 from https://github.com/amplab/tachyon/releases/tag/v0.4.1

• Compile sparkling demo

cd h2o-sparkling-demo
sbt assembly

Note: The assembly stage is important, since the demo is a Spark driver sending a jar-file containing implementation of a working job.

For this run no Spark cloud is required:

• Execute an instance of H2O embedding Spark driver

cd h2o-sparkling-demo
sbt "run --local"

For this run a Spark cloud is required:

• Run master and one worker on local node

cd spark/sbin
./start-master.sh
./start-slave.sh 1 "spark://localhost:7077"

• Assembly h2o-sparkling-demo jar file which can be sent by the driver to Spark cloud

cd h2o-sparkling-demo
sbt assembly
sbt "run --remote"

cd h2o-sparkling-demo
sbt runH2O

Currently demo supports three extractors:

• dummy - pull all data into driver and create a frame
• file - ask Spark to save RDD as a file on local filesystem and then parse a stored file
• tachyon - ask Spark to save RDD to tachyon filesystem, then H2O load a file from tachyon FS

The extractor can be selected via --extractor command line parameter, e.g., --extractor==tachyon

• Start Tachyon

cd tachyon/bin
./tachyon-start.sh

• Look at http://localhost:19999/ to see list of files stored on the storage or type tfs command tachyon tfs ls /

• For more info details discuss instructions instructions on http://tachyon-project.org/Running-Spark-on-Tachyon.html

Run a demo with Tachyon-based extractor againts remote Spark cloud:

cd h2o-sparkling-demo
sbt assembly
sbt "run --remote --extractor=tachyon"

Run airlines demo with file-based extractor againts remote Spark cloud running on non-default location:

sbt "run --remote --sparkMaster=spark://localhost:17077 --noshutdown --demo=airlines --extractor=file"

• Matei Zaharia' slides about Spark
  • http://ampcamp.berkeley.edu/wp-content/uploads/2012/06/matei-zaharia-part-2-amp-camp-2012-standalone-programs.pdf