Software installation on a High Performance Cluster (HPC) is different from the usual software installation on a desktop computer.
In order to get the best performance out of the very expensive hardware, it is best to compile software from source, rather than using pre-compiled binaries. This sounds often easier than it actually is.
One good example is Basic Linear Algebra Subprograms (BLAS) [1]: A generic installation will only deliver a very low performance. Optimising the compile process will deliver more performance. However, with the sheer amount of different CPUs around, testing it for each new build with a new compiler is time consuming.
The solution here is to fall back what others have done, improve on it if possible and contribute back to the community.
One such an example is EasyBuild [2]: software which is installing high-performant scientific software on HPC clusters automatically.
The talk will take you through an often a bit personally journey of scientific software installation, without too much of technical jargon. It is aimed for scientists who are installing software, or write their own code. Equally, researchers who are more experienced in software installation might learn something new as well.
Speaker
Jörg Saßmannshausen is a fully trained Chemist who gained his PhD at the University of Leeds (UK) in 1997. Up until 2010, he was working at various research institutions (Oxford, Imperial, Max-Planck-Institut in Germany, University of Strathclyde), and gained his venia legendi from the Technical University in Graz/Austria in 2011. In 2010 he switched from Chemistry as a daytime job and IT on the side to be a full time IT support and worked at various research institutes in London (UCL, The Crick, GSTT). He is currently working for Imperial College London.
That said, he is still a gentleman scientist with an active research portfolie.
Further details can be found here: http://sassy.formativ.net/