Skip to content
Go to file

Latest commit


Git stats


Failed to load latest commit information.
Latest commit message
Commit time


@Copyright 2013-2018 Inidana University Apache License 2.0


Harp is a HPC-ABDS (High Performance Computing Enhanced Apache Big Data Stack) framework aiming to provide distributed machine learning and other data intensive applications.


  1. Plug into Hadoop ecosystem.
  2. Rich computation models for different machine learning/data intensive applications
  3. MPI-like Collective Communication operations
  4. High performance native kernels supporting many-core processors (e.g., Intel Xeon and Xeon Phi)

Online Documentation

Please find the full documentation of Harp at including quick start, programming guide, and examples.

Open Channels

  1. Google group -
  2. Slack -

Installation of Harp

Install from Binaries

Please download the binaries of Harp from

Copy the jar files to $HADOOP_HOME

## the core modules 
cp core/harp-hadoop-0.1.0.jar $HADOOP_HOME/share/hadoop/mapreduce/
cp core/harp-collective-0.1.0.jar $HADOOP_HOME/share/hadoop/mapreduce/
cp core/harp-daal-interface-0.1.0.jar $HADOOP_HOME/share/hadoop/mapreduce/
## the application modules 
cp ml/harp-java-0.1.0.jar $HADOOP_HOME/
cp ml/harp-daal-0.1.0.jar $HADOOP_HOME/
cp contrib-0.1.0.jar $HADOOP_HOME/ 

Install from Source Code

  1. Install Maven by following the maven official instruction
  2. Compile harp by Maven with different hadoop versions
## x.x.x could be 2.6.0, 2.7.5, and 2.9.0 
mvn clean package -Phadoop-x.x.x
  1. Copy compiled modules jar files to $HADOOP_HOME
cd harp/
## the core modules 
cp core/harp-hadoop/target/harp-hadoop-0.1.0.jar $HADOOP_HOME/share/hadoop/mapreduce/
cp core/harp-collective/target/harp-collective-0.1.0.jar $HADOOP_HOME/share/hadoop/mapreduce/
cp core/harp-daal-interface/target/harp-daal-interface-0.1.0.jar $HADOOP_HOME/share/hadoop/mapreduce/
## the application modules 
cp ml/java/target/harp-java-0.1.0.jar $HADOOP_HOME/
cp ml/daal/target/harp-daal-0.1.0.jar $HADOOP_HOME/
cp contrib/target/contrib-0.1.0.jar $HADOOP_HOME/ 

Add third party dependencies

Harp depends on a group of third party libraries. Make sure to install them before launching the applications

cd third_party/
## JAR files
cp *.jar $HADOOP_HOME/share/hadoop/mapreduce/
## DAAL 2018
## copy daal java API lib
cp daal-2018/lib/daal.jar $HADOOP_HOME/share/hadoop/mapreduce/
## copy native libs to HDFS
hdfs dfs -mkdir -p /Hadoop
hdfs dfs -mkdir -p /Hadoop/Libraries
hdfs dfs -put daal-2018/lib/intel64_lin/ /Hadoop/Libraries
hdfs dfs -put tbb/lib/intel64_lin/gcc4.4/libtbb* /Hadoop/Libraries

Installation of Harp-DAAL-Experimental (from source code)

Harp-DAAL-Experimental only supports an installation from source code for now. Please follow the steps

  1. Pull the DAAL source code branch: daal_2018 branch
git clone -b daal_2018
mv harp harp-daal-exp
cd harp-daal-exp

or git pull the submodule from third_party/daal-exp/

cd harp/
git submodule update --init --recursive
cd third_party/daal-exp/
  1. Compile the native library either by icc or gnu
## use COMPILER=gun if icc is not available
make daal PLAT=lnx32e COMPILER=icc
  1. Setup DAALROOT environment variable by sourcing scripts from DAAL release codes.
source ../__release_lnx/daal/bin/ intel64
  1. Compile harp-daal-experimental modules at Harp. Makesure that line 17 of harp/pom.xml file is uncommented and DAALROOT is setup by step 3.
### check DAALROOT
### re-run maven to compile
mvn clean package -Phadoop-x.x.x
  1. Install compiled libraries.
## copy Java API to Hadoop folder
cp ../__release_lnx/daal/lib/daal.jar $HADOOP_HOME/share/hadoop/mapreduce/
## copy harp-daal-exp libs
cp experimental/target/experimental-0.1.0.jar $HADOOP_HOME/
## copy native libs to HDFS 
hdfs dfs -mkdir -p /Hadoop
hdfs dfs -mkdir -p /Hadoop/Libraries
hdfs dfs -put ../__release_lnx/daal/lib/intel64_lin/ /Hadoop/Libraries
hdfs dfs -put ../__release_lnx/tbb/lib/intel64_lin/gcc4.4/libtbb* /Hadoop/Libraries
hdfs dfs -put harp/third_party/omp/ /Hadoop/Libraries/
hdfs dfs -put harp/third_party/hdfs/* /Hadoop/Libraries/

The experimental codes have only been tested on Linux 64 bit platforme with Intel icc compiler and GNU compiler.

Run example of K-means

Make sure that harp-java-0.1.0.jar has been copied to $HADOOP_HOME. Start the Hadoop service


The usage of K-means is

hadoop jar harp-java-0.1.0.jar edu.iu.kmeans.regroupallgather.KMeansLauncher
  <num of points> <num of centroids> <vector size> <num of point files per worker>
  <number of map tasks> <num threads> <number of iteration> <work dir> <local points dir>

For example:

hadoop jar harp-java-0.1.0.jar edu.iu.kmeans.regroupallgather.KMeansLauncher 1000 10 100 5 2 2 10 /kmeans /tmp/kmeans
You can’t perform that action at this time.