Monday, May 30, 2011

Will next generation of high performance computing and clouds go the way of telescopes or particle accelerators?

Will next generation of  super computers go the way  of telescopes or particle accelerators?

My talk at TICAL 2011

Years ago many universities had their own research telescopes and small accelerators. But as the demands of science increased, as well as the costs, researchers quickly realized they had to consolidate their resources and build instruments that served the needs of hundreds or thousands of researchers around the globe.  Virtually of all today’s big science instruments such as telescopes, particle accelerators, synchrotrons, etc are multi-country collaborations.  Research computing may be headed in the same direction. The next generation of super-computers and research cloud infrastructure required for climate modeling, weather forecasting, epidemiological studies, etc will require massive amount of electrical energy to operate. The energy costs alone may compel international partnership to deploy and build such infrastructure on the same scale of global collaboration as we have seen for telescopes and particle accelerators.  More importantly with the growing threat of climate change it is critical that such facilities not be major sources of CO2 emissions in their own right as exemplified by the new climate modeling super computer in Exeter in the UK and the recently constructed NCAR data center in Wyoming.  We are already seeing early signs of such research computing collaborations as for example the investigation by CERN to relocate its data center to Nordic countries and universities in the Boston area to relocate their computing facilities to a small municipal hydro-electric facility 90 miles west of Boston.  Such research computing collaborations will also significantly save individual universities millions of dollars in electrical costs as currently research computing represents 15-30% of the electricity consumption at many universities.  The energy savings alone could possibly pay for this next generation of research computing and still leave additional money to support critical research. Obviously high speed optical networks and open lightpath exchanges will be critical to such a reality. But it is just as important that energy and environmental savings not be transferred to the higher costs in the network and so new low carbon network architectures are needed as well.