sara-nl/lsg-rbuilder

Name: lsg-rbuilder

Owner: SURFsara

Description: Scripts to build optimized versions of R on the SURFsara Life Sciences Grid

Created: 2014-08-21 12:13:07.0

Updated: 2017-11-28 17:19:14.0

Pushed: 2014-08-26 10:02:42.0

Homepage: null

Size: 160

Language: Shell

GitHub Committers

UserMost Recent Commit# Commits

Other Committers

UserEmailMost Recent Commit# Commits

README

Optimized OpenBLAS and R on the SURFsara Life Sciences Grid

Introduction

The Life Sciences Grid has a basic set of packages supplied by Centos. These packages are not optimized for the architecture of the compute nodes of the life-sciences grid. For better performance, you need to compile optimized versions of the software of choice yourself.

OpenBLAS can be optimized for the Bulldozer architecture on the Life Science Grid compute nodes. If R has to be compiled against an optimzied version of OpenBLAS, it must be compiled on the architecture it was optimized for. This means that R cannot be compiled on the user interface machines, because the user interfaces are virtual machines, which have a limited CPU instruction set.

The solution is to compile OpenBLAS and R on the worker nodes, within a job.

Compiling OpenBLAS and R

This directory contains a script which downloads and compiles OpenBLAS and R. It uses the most recent versions available at the time of writing:

The script should not be run on the user interface machines. Instead, run the script as a grid job:

qsub R-3.1.1-build-job.sh

This will download OpenBLAS and R, and install everything in

$HOME/opt/R-3.1.1

Running Benchmarks

In the 'benchmarks' directory, there are a few R-scrips for running benchmarks mentioned on the sites below. To run the benchmarks, first, downloade the benchmark tests from sites below. Then submit the jobs using

qsub -q express mass-bench-dist.sh

and

qsub -q express mass-bench-openblas.sh

References:

  1. http://r.research.att.com/benchmarks/
  2. http://www.cybaea.net/Blogs/Faster-R-through-better-BLAS.html

This work is supported by the National Institutes of Health's National Center for Advancing Translational Sciences, Grant Number U24TR002306. This work is solely the responsibility of the creators and does not necessarily represent the official views of the National Institutes of Health.