What does successful industrial collaboration look like?

I sat in on an interesting session at the CASC-HPCSIG meeting in Oxford last week, looking at different models for university-industry cooperation in high-performance computing. All considered that people, support and expertise are at least as important to a successful liaison as processors, so were slightly puzzled that publicity, bids and even informal discussions tend to focus almost entirely on size of hardware.

The Texas Advanced Computing Centre now gets more income from sharing expertise with local industries than selling machine time to them. Large industries (oil, obviously!) have their own high-performance computers but are very interested in subscribing to TACC's affiliate programme where they can compare notes with peers from both the university and their competitors. This includes quick phonecalls to find out how new software was installed, formal workshops to discuss common issues, or testing production code on new hardware. Real geological data is seen as the prize asset, not to be placed on anyone else's computer, so for testing a company created a realistic model oilfield that university students can also use for their projects. Smaller companies may want to use the centre's equipment but are most interested in help with modelling: how to use fluid dynamics software or to tune their code for parallel processing.

Concern was expressed at any assumption that relationships between universities and industry were one-way, with universities providing equipment and expertise and companies providing money. Success requires much more complex interactions, and needs to recognise the different requirements of the two parties. Industry partners are likely to have a specific problem they need resolved, to a tight timescale and budget. It was suggested that liaison units are more successful when run by people with a background in business, even sales. Universities may well need to behave as suppliers into a (crowded) market rather than partners in a research endeavour.

All the speakers agreed that an "HPC Centre" should be something with expert people and support mechanisms, not just a collection of processors in racks (though the latter is much easier to put into glossy brochures). Which led to the interesting question, whether the measure of a HPC Centre's success should be that industry use of its CPUs is increasing? Or decreasing, because it has helped industry develop the skills and expertise to design and use their own?