The key factors in most outsourcing decisions are cost and availability. Many universities and other research organizations will indeed consider outsourcing of HPC services, i.e., acquiring research computing cycles and services from private sector providers. They would already be doing so to a larger extent today, but it is the cost considerations that prevent them from going in that direction. Several major companies, including IBM, Sun, Amazon and others, have had product offerings in this space — cycles or cycles-plus-services — on the market for a couple of years now. But that part of their business has not seen big growth.
Cost is the issue. Universities find it more cost-effective, even after taking into account the increased investment in infrastructure and doing as thorough a total-cost-of-ownership calculation as possible, to build and operate their own HPC services than to buy such computing services from industry. The cost structure for the U.S.-based companies makes them non-competitive thus far as a provider of such services.
There are no major advertisement dollars in large-scale computations for research and development work. Revenue from targeted advertisements is a key driver in the outsourcing of e-mail service, despite legitimate security and privacy related concerns on the part of academic institutions. The same type of shifting of cost away from universities, and by way of selling access to consumers, on to the industry providers, will likely happen in other areas of academic IT services. But the key distinction is that e-mail and other such services, although quite necessary and indeed vital, really don’t offer one academic institution major competitive advantage over another. But their research infrastructure surely does. There are measurable benefits in having a non-commodity and competitive research computation infrastructure. This is true for both academia and industrial R&D.
Even if the U.S.-based private sector companies could significantly alter their cost structure and be able to offer more competitive products, at least for raw compute cycles and basic system services, it would still be hard for them to compete today because of the very substantial role the federal agencies have. It is hard for industry providers to compete with “free,” i.e., when national centers and labs, supported by various federal grants, are offering their cycles, for good reasons, at no direct cost to the end-users. If the federal government agencies were to collectively alter their HPC resource provisioning policies and instead offer funding to researchers to buy cycles and services on the open market, there is a possibility that a real market may begin to emerge — a market where private sector companies, national centers and the university-based IT service organizations can compete on a more level playing field. The real consumers, the end-user scientists and engineers, can then pick what is most cost-effective for their particular set of needs in terms of cycles and services. They can then make better choices by taking cost considerations into account.
If there was significantly more demand from both the academic and industrial R&D consumers, it is possible that a more robust compute services industry segment can emerge. It is possible that the way in which federal agencies currently fund research computation related cyberinfrastructure has much to do with the emergence of a large compute services industry that is primarily based in the United States. This industry will not only offer on-demand compute capacity as in “cloud computing,” but also value-added simulation and analysis services, an area in which U.S.-based companies, in stronger partnership with academia, can be globally competitive.
If we open up to the possibility of compute infrastructure going truly global — and that means services coming from providers based in Eastern Europe or the Far East (India, China and so on) — it raises several policy questions: Are the U.S. academic institutions ready and should they be open to buying compute and data cycles from overseas providers? Are there security issues and should or can the U.S. federal R&D funding be spent on acquiring services from non-U.S. companies? It raises a whole new set of questions, few of which the academic community has thought through yet.
Many sectors of U.S. industry have worked through the issues of separating design and the research and innovation that goes into it from manufacturing and fabrication. They are comfortable with many of the chip fabs being in the Far East. They have processes in place to take care of all related issues. The academic community should also think through its own set of challenges and opportunities in this space.