Skip to content

Latest commit

 

History

History
116 lines (104 loc) · 8.99 KB

README.md

File metadata and controls

116 lines (104 loc) · 8.99 KB
Raw

Arkouda-NJIT

This is an external repository to build functionality for Arkouda with a focus on advanced graph processing. It is built with the same structure as arkouda-contrib to manage modules and easily swap between the production (arachne) and development (arachne_development) directories.

Configuring Prerequisites

To install the prerequisites below, the following libraries must be installed on your system. This can be done via any package handler depending on your distribution, or if on a cluster HPC system, they can be loaded in as modules. At the time of writing, the following versions were confirmed to work.

  1. GCC 11.2.0 or later.
  2. CMake 3.26.3.
  3. OpenMPI GNU 4.1.4 (needed by CMake)
  4. Anaconda 2023.09-0
  5. jq 1.6 command-line JSON processor.

Prerequisite Installation Steps

We recommend following the installation instructions provided by the Arkouda development team. Most specifically, follow the Prerequisites section in its entirety, and only the Dependency Configuration section of the build instructions. The installation steps usually involve the following:

  1. Download Chapel from the Chapel downloads page. Use Chapel version 2.1.0.
    • Alternatively, you may clone Chapel and switch to a given tagged version. The commands for these should look something like: 
      git clone https://github.com/chapel-lang/chapel.git
cd chapel
git fetch --tags origin
git checkout tags/2.1.0 --force
  2. Build Chapel by executing the commands below. This assumes you have installed all Chapel prerequisites. Note: We recommend using gcc/11.2.0 or later due to dependencies with VieCut. 
    cd /path/to/chapel/
source ./util/setchplenv.bash
export CHPL_GMP=bundled
export CHPL_HWLOC=bundled
export CHPL_RE2=bundled
export CHPL_LLVM=bundled
make -j 4 # This value can be increased dependent on your device's number of processors
    • Note: This installs single locale (shared-memory) Chapel. For multilocale (distributed-memory) Chapel please follow the documentation guide on Multilocale Chapel Execution. Arachne has its best performance on shared-memory. However, kernels such as breadth-first search and property graph querying have multilocale-optimized versions that require multilocale Chapel to be installed.
  3. Download, but do not build, Arkouda. Use Arkouda version v2024.06.21. A specified version can be selected for download by clicking on Releases in the main GitHub page for Arkouda.
    • Alternatively, you may clone Arkouda and switch to a given tagged version. 
      git clone https://github.com/Bears-R-Us/arkouda.git
cd arkouda
git fetch --tags origin
git checkout tags/v2024.06.21 --force
  4. Install Arkouda dependencies with Anaconda. An environment containing all dependencies can be installed from arkouda-env.yml or arkouda-env-dev.yml within your Arkouda home directory.
    • This can be done by executing the following command within your Arkouda directory: 
      conda env create -f arkouda-env-dev.yml
  5. Configure your Arkouda dependencies. This involves creating (or modifying) the Makefile.paths within your Arkouda home directory.
  6. Install constrained-clustering and compile the C++ object files required by Arachne by following the commands below. Constrained-clustering requires a C++ compiler that supports c++-20, such as clang++11, and cmake. These and other prerequisites should be covered by the prerequisites in items 1-5 above. 
    cd /path/to/arkouda-njit/arachne/server/external_libs
git clone https://github.com/MinhyukPark/constrained-clustering.git
cd constrained-clustering
./setup.sh
./easy_build_and_compile.sh
cd ../../../../arachne/server/viecut_helpers/
source compileLogger.sh -f logger.cpp -o logger.cpp.o
gcc -c -fPIC -I../external_libs/constrained-clustering/external_libs/VieCut/lib/ -I../external_libs/constrained-clustering/external_libs/VieCut/extlib/tlx/ computeMinCut.cpp -o computeMinCut.o
cd ../../../

Building Arachne

Building Arachne is performed through executing the module_configuration.py file. The complete path to the location of arkouda must be specified through ak_loc and the complete path to the location of arachne should be specified through pkg_path.

python module_configuration.py --ak_loc=/complete/path/to/arkouda/ --pkg_path=/complete/path/to/arkouda-njit/arachne/ | bash

The above command will pipe the following three commands to terminal that installs Arachne using pip, copies the Arkouda server modules to a temporary file, and combines them with the Arachne server modules to build the enhanced arkouda_server.

pip install -U /complete/path/to/arkouda-njit/arachne/client
cp /complete/path/to/arkouda/ServerModules.cfg ~/TmpServerModules.cfg.1683320760
ARKOUDA_SERVER_USER_MODULES=" /complete/path/to/arkouda-njit/arachne/server/BuildGraphMsg.chpl  /complete/path/to/arkouda-njit/arachne/server/PropertyGraphMsg.chpl  /complete/path/to/arkouda-njit/arachne/server/GraphInfoMsg.chpl  /complete/path/to/arkouda-njit/arachne/server/BFSMsg.chpl  /complete/path/to/arkouda-njit/arachne/server/TriCtrMsg.chpl  /complete/path/to/arkouda-njit/arachne/server/TriCntMsg.chpl  /complete/path/to/arkouda-njit/arachne/server/TrussMsg.chpl  /complete/path/to/arkouda-njit/arachne/server/CCMsg.chpl" ARKOUDA_CONFIG_FILE=~/TmpServerModules.cfg.1683320760 ARKOUDA_SKIP_CHECK_DEPS=true make -C /Users/alvaradoo/Research/arkouda

For usage instructions of module_configuration.py please execute the the following.

python module_configuration.py --help

Building Development Arachne

If you are interested in installing the development version of Arachne, please follow the same instructions as above, but for pkg_path include /complete/path/to/arkouda-njit/arachne_development/.

Starting the Arachne and Arkouda Server

The server can be started as specified in the Arkouda documentation. Simply put, navigate to your Arkouda directory, and an executable named arkouda_server should exist. Execute it with the command below to start a server instance.

./arkouda_server # - nl X

The output should be something that looks like below.

********************************************************************************************************
********************************************************************************************************
*                                                                                                      *
*                          server listening on tcp://n118.cluster.local:5555                           *
*                                 arkouda server version = v2024.06.21                                 *
*                                    built with chapel version2.1.0                                    *
*                                     memory limit = 973796998348                                      *
*                                       bytes of memory used = 0                                       *
*                                                                                                      *
********************************************************************************************************
********************************************************************************************************

To run a simple test file as well as pytests please proceed to the Arachne directory for those instructions.

Usage Notes

import arkouda as ak
import arachne as ar
# code using arachne and arkouda below

Common Issues

  • Issue: Unrecognized HDF5, Apache Arrow, etc. installations. Fix: Ensure Makefile.paths was properly added to the base Arkouda directory. More information can be found in the Arkouda build instructions.
  • Issue: Arkouda or Arachne functions are not recognized when executing scripts. Fix: Make sure to run pip3 install -e . at both /complete/path/to/arkouda-njit/arachne/client/. and /complete/path/to/arkouda/.