Usage ideas for a cluster at a university [closed]

you could install debian plus the various debian-science packages on these machines. that includes quite a large collection of free software programs for several common computation problems in science, as well as general libraries and toolkits such as openmpi.

see also http://blends.alioth.debian.org/science/tasks/. it contains a list of the sub-projects or Tasks of the DebianScience project as well as listings of the software packages available, with descriptions of what they do/are for, and the state of the packages. The main D-S wiki page above has a broken link to this page, so here's a fixed link.

I currently work as sysadmin in the school of chemistry of a university here in .au, and have recently been building several machines for the academics with the DebianScience/Chemistry packages installed plus commercial/proprietary software such as Gaussian03, QCHEM, and WebMO (which is a web/java front-end to Gaussian, QCHEM, MOPAC, and other computational chemistry programs).

I'm more familiar with the Chemistry programs in the Debian Science packages, but i know there's also a huge amount of software for other fields such as Physics, Astronomy, Biology, Mathematics, and so on. Also more "general" packages for data aquisition, typesetting (TeX, etc), computation libraries for Fortran, C, python, and more.

if nothing else, that debian-science wiki page above will give you a good overview of the kinds of software that science academics might be interested in on computing clusters. You said you have three groups of machines for clusters, so once you've got the overview on what kinds of software are available you can talk to some of the professors and start making plans for rebuiling/re-purposing the machines.

When I was sysadmin for a University lab I was faced with a vast amount of processing power (around 50 iMacs) that was largely underused, so wanted to do the same thing that you're suggesting. To gain the initial traction I found a PhD student who had some parallelisable problem - this was Physics, so he had a Lattice QCD simulation - and set about porting his code to the Mac so it would run under Xgrid. When he got results back in 1/3 the time it would have taken on the shared Sun cluster, on what was effectively "free" CPU time, the rest of the department paid attention.

A hadoop cluster could be used to process massive amounts of data if they have the need.