Setting up my own computing cluster with a batch queuing system and then using it run large numbers of molecular dynamics simulations was one of the more satisfying things I have done professionally. The compute nodes were Apple Xserves from 2008 and 2009. Myself and Ben Hall won the first seven of these nodes from the Apple Research and Technology Support (ARTS) programme ten years ago. Ben used them for a while and named them glados, after the AI in portal, so I kept the name when I retrieved the nodes when I moved to the John Radcliffe.
Alas, these nodes are now too slow, and therefore inefficient, compared to modern machines. For example, each CPU core in the 40-core nodes that we were able to purchase as part of the recent NIHR grant are 4.7x faster when running GROMACS than each of the 16 cores in each glados node. Ignoring scaling, this means that all 16 nodes of glados are equivalent to just one and a third of these new nodes, taking up 12x less rack space.
So, goodbye glados!
