You will need:

• A Linux system recent enough to have perf-events
• A CUDA installation (>= 6.0)
• A CMake installation for building oskar_binary

Quickest way to get going is to install the Haskell Plattform (http://haskell.org/platform/). But you can also obtain a binary Haskell package from http://www.haskell.org/ghc/dist/7.8.3/

 configure --prefix $HOME/opt
 make install

You will also need cabal-install from http://www.haskell.org/cabal/download.html:

 install the binary in your path

For building you first need an installation of oskar_binary, which you might have to request from the OSKAR development team (http://www.oerc.ox.ac.uk/~ska/oskar2/):

unzip oskar_binary.zip
cd oskar_binary
cmake .
make

Then check out the code and build:

 git clone https://github.com/SKA-ScienceDataProcessor/RC.git
 cd RC/MS3
 sh boot.sh
 cabal install

A number of programs should have been installed into bin/. For example, a number of different distributed dot product versions can be started using e.g.:

bin/ddp-in-memory --nprocs <N>
bin/ddp-in-memory-collector --nprocs <N>
bin/ddp-in-memory-hierarchical --nprocs <N>

Replace "" with the number of cores you would like to use. Note that the "hierarchical" version is about testng hierarchical failure propagation, so it is expected for the program to fail.

The gridder is meant to be run as a Cluster application in a SLURM environment. It expects OSKAR visibility files with names of the form test_p%02d_s%02d_f%02d.vis to be present in the working directory. It will require 21 nodes to run through.

The full command line should be as follows:

/absolute/path/to/RC/MS3/bin/Gridder_DNA

The imaging program generates automatically generates profiling information into $HOME/_dna. For generating a profile overview first make sure to install the command line ghc-events profile reader:

cabal install ghc-events

Then switch to the visualize sub-directory for MS4 and invoke the generator for the run, for example:

cd visualize
python html_plotter.py timeline ~/_dna/1634160-s 1634160-s.html

Which should generate a profiling report.