Overview

Taproot contains code to evaluate the creation of a software stack for performing scientific data analysis. The stack can currently be built to evaluate anlyzing scientific data with various packages. Some of those packages require complex build system configurations.

The available analysis stacks are:

Apache Lucene
Lawrence Berkeley Lab's Fastbit
LevelDB
Apache Arrow and Parquet
DuckDB
TrinoDB

Taproot Quickstart Guide

Just Build Taproot

Assuming you have everything else configured, maybe you just want to build taproot.

Do this: Edit gradle.properties to point at the install path for components

./gradlew build
mkdir build
cd build
cmake ..
cmake --build .

Just run Taproot

cd data/1m_points
gzip *
cd ../../build
mfem2parquet -s 0 ../data/1m_points test.parquet
duckdb test.parquet

General Prerequisites and Taproot Dependencies

If you haven't built Taproot before, you'll need to access several supporting packages to generate data, etc.

You will need to either build Laghos and all of its pre-requisites to generate a Laghos-formatted MFEM mesh, or if you already have a Laghos mesh stored in files you can simply build a serial version of MFEM and use that to construct arrays of Laghos mesh data.

Building MFEM for use with previously generated mesh data

• Note that to use MFEM with Lucene it is necessary to build MFEM so that it can
• be linked dynamically. By default these steps do not occur.

  wget -c https://bit.ly/mfem-4-5 -O mfem-4.5.tgz
  tar xvfz mfem-4.5.tgz
  cd mfem-4.5
  edit config/defaults.mk PREFIX=${SWHOME}, SHARED=yes
  make config BUILD_DIR=build
  cd build
  make -j 4 serial
  make PREFIX=$SWHOME SHARED=YES install
  make STATIC=YES PREFIX=$SWHOME install

NOTE: The following CMake steps do not work with this version of MFEM.

  CMake installation does not work, these instructions will not enable -fPIC
  mkdir build
  cd build
  cmake -DCMAKE_INSTALL_PREFIX=$swhome -DCMAKE_INSTALL_RPATH=$swhome/lib -DMFEM_USE_MPI=no ..
  make install

Build Apache Arrow C++ with Parquet support

  git clone [email protected]:apache/arrow.git
  cd arrow
  git checkout apache-arrow-13.0.0
  cd cpp
  mkdir -p build
  cd build
  cmake .. --preset ninja-debug -DARROW_BUILD_STATIC=ON -DARROW_BUILD_EXAMPLES=ON -DPARQUET_BUILD_EXECUTABLES=ON
  cmake --build . -j 2
  cmake --install . --prefix $SWHOME

Building Apache Lucene Analysis Capability

The following steps describe how to build the taproot Lucene data analytics package

Build and Install Gradle

Install curl

curl -s "https://get.sdkman.io" | bash
sdk install gradle 8.1.1

Download and Build Apache Lucene version 10.0

Install OpenJDK 17 or 18 (on ubuntu, usually update-alternatives --config java is sufficient)

git clone https://github.com/apache/lucene
cd lucene
./gradlew

Build and Install TrinoDB

Other planned support

Fastbit and TrinoDB support are not yet enabled.

Building Taproot Components

Configure Taproot to use Lucene Indexing capability

cd taproot Edit gradle.properties to set the path to MFEM and Arrow

Build Taproot Lucene MFEM Utils

  ./gradlew build
  ./gradlew runTest

Build Taproot Lucene for Science

  ./gradlew build
  ./gradlew test

Run Taproot with Lucene for Science

• Stop all existing gradle daemons

  ./gradlew --stop

• Even though we specify no daemon, one daemon runs and exits at shutdown (we'd prefer no Daemon at all)

  ./gradlew run --no-daemon

Build the Taproot Parquet

  sudo apt-get install libboost-all-dev
  cd TAPROOT_DIR
  mkdir build
  cd build
  cmake .. -DARROW_DIR=$SWHOME -DMFEM_DIR=$SWHOME
  cmake --build . -j 1

Converting MFEM data to Parquet

  $SWHOME/bin/mfem2parquet -s flat|custom|trino mfem_dir parquet_dir

Generate a fastbit index

TODO: ./taproot-create-fastbit-index -i <mesh_dir> -o TODO: ./taproot-select-fastbit -i -q

Generate a KDTree index