High Performance Computing (HPC) enables advanced scientific calculation, simulation and modelling, which in turn, and to an increasing extent, is a precondition for much of the research and innovation that is fundamental for today's knowledge driven economy. The Scandinavian countries spend millions of Euros every year on supercomputers and their electricity consumption.
"Supercomputing has become fundamental for science and innovation, yet when the cost for hosting and operations is becoming comparable to the costs of hardware, and investments are increasing, we need to look into cost efficient solutions", stated Jacko Koster, director of UNINETT Sigma.
Added to this economic incentive is of course the environmental one. Supercomputers entail a large CO2-footprint when fossil energy sources are used. In Iceland, energy is produced not only at low cost but also from CO2-neutral renewable hydro- and geothermal energy sources. Due to Iceland's geographical location, it is not feasible to transfer electricity to Europe. Hardware, however, can be moved, and so can data, via the trans-Atlantic fibre optic data-network infrastructure.
In the long term, joint large scale procurements and energy efficient placement of supercomputers will be increasingly advantageous for the Scandinavian countries as well as to Iceland. It increases value for money as well as the possibility to develop new advanced competencies within shared operations of remote computing.
"We need to constantly develop our understanding of advanced computing and how to operate it in increasingly complex ways", stated Ebba Þóra Hvannberg, director of the project and of Icelandic Supercomputing. "We must continuously push the total cost of ownership down and increase the value for money", added Rene Belso, director of Danish Supercomputing, continuing: "Indeed, we Nordics need to be first movers in all such area, since we only seem to be able make a national business case out of the most complex organisation and advanced technology implementations." Like with many other technology fields, e.g. environmental technologies, early public piloting can make the Nordics world leaders in related commercial fields.
The project is the result of collaboration between the Danish Center for Scientific Computing (DCSC), the
Swedish National Infrastructure for Computing (SNIC), UNINETT Sigma and the University of Iceland. The
compute facility will be hosted by Thor Data Center, now part of the Advania family of enterprises.
Jacko Koster said that "If the pilot project is successful, successor projects may be defined in the coming years, e.g., for the joint procurement of larger supercomputers or specialized systems, which one country cannot afford alone. Possibly, such Nordic infrastructure can also be a joint contribution to European Infrastructure, like that of the Partnership for Advanced Computing in Europe (PRACE)".
Such ideas are, understandably, not always shared by university computer centres, presently hosting the
supercomputers. Indeed, they often argue the necessity of having the hardware close by, even at higher
operations costs. "We do understand the concern of traditional computer centres, but maintain that they also regularly need to review their operations strategy, and think of cost efficiency. There will be a continuous need for complex supercomputing requiring close attention, experimentation and constant tweaking. Therefore, Nordic computer centres should focus on advanced operations and user support, not on hardware maintenance", stated Rene Belso.
Sverker Holmgren added: "Eventually, the aim is that national infrastructures for computational science in the participating countries can further increase focus on delivering high quality services and access to computational infrastructures for their users, whereas the more elementary aspects of the infrastructure (e.g., hosting of equipment) could be handed over to parties that can implement this in a more cost efficient manner, without compromising quality of service."
"Knowing that the project already consists of many complexities of a political, organisational and administrative nature, we aimed for a robust standard supercomputer architecture, useful to most researchers", stated Ebba Þóra Hvannberg. The system is being delivered by HP, via Opin Kerfi. It is based on a cluster of 288 HP ProLiant BL280c G6 servers with 3456 compute cores, achieving 35 TeraFLOPS of peak performance. Additionally it includes a 72 terabyte HP IBRIX X9320 Storage system. Project management, installation, implementation and the testing of equipment was the responsibility of the Opin Kerfi. The solution fulfilled all the requirements set out in the project scope. More information is available at http://nhpc.hi.is .