Talked to the Stuttgart HPC centre. Of particular interest was the use of visualisation of OpenFOAM simulation (in this case to determine the best way for a cyclist to sit on a bicycle) using 3D glasses. This could be useful in many contexts in many universities, not just sport. It was a huge screen, though, so quite an investment so perhaps that’s the sort of thing that would be better at a smaller scale (Occulus rift, etc?). But a large screen does allow more than one person to collaborate effectively.
I saw the presentation on the purely water-cooled data centre at Birmingham. This has an impressive PUE of 1.03, using a large amount of free cooling (all but a few days a year) and with lots of good statistics.
I talked to Rescale. This offers a US-government approved multi-cloud brokerage system. One thing in particular that I discussed was billing models. For example, where there is a budget of a fixed number of core hours then Rescale supports a fixed alert (say 90% usage) and a fixed cut off, which would allow spend on the public cloud side to be controlled, but in reality decisions on allowing a group additional core hours to complete work have a time delay, e.g. a couple of weeks. So I suggested that a better option would be determining spend rate and giving an alarm when the time is likely to be used up within, say, a two week period in addition to other alerts so the process of getting additional time can be started soon enough. They seemed to take this on board. With a decent API something external to alert could also be written, of course. Whilst external cloud has some cost issues (although the figures depend on where you draw the line on costing of on-premise), there are definitely use cases where it makes sense, e.g. bursting (if the costs and returns can be controlled) or perhaps more obviously the ability to burst out to provide teaching resources beyond an existing on-premise capacity, provided there are the funds to do this (and that’s another policy and financial issue to be sorted, and in conjunction with timetabling as shifting the demand for computational resources for teaching on a course could save quite a lot of money if it meant that the on-premise resource could cope with the demand). It could also allow flexibility on what resources are made available which is a benefit over relaively fixed on-premise provision, but it’s notoriously hard to determine what the value of that benefit is compared to the cost.
The exhibitors area was very crowded but I will be back later to talk to Bios IT and others who made approaches as well as talk to others that seem to have things on offer, There were certainly a lot of what seemed to be young companies offering immersive cooling solutions and various storage solutions such as WEKA. Determining the true value being offered, or longevity of those organisations is not trivial.