SaaS has acquired a new meaning, Supercomputing as a Service, thanks to Cambridge University’s desire to turn its High Performance Computing (HPC) Solution Centre into a profit centre from being a cost burden. The answer has been to turn it into a cloud-delivered SaaS operation.
The Centre, the result of a partnership between the University’s School of Physical Sciences and computer maker, Dell, was formally opened this week. It now fronts much of the work undertaken by the Darwin cluster, which is based on commodity Dell servers. These have recently been upgraded to systems running the latest Intel Nehalem processors.
Darwin is regularly used by over 400 researchers is some 70 different research groups. To date these have been almost exclusively in the pure and applied sciences area, but the Centre is now starting to see work appearing in the areas of social sciences and financial services, particularly in disciplines such as risk analysis. It is now also starting to attract interest from potential users outside of the University research community.
This is being accelerated by the availability of a cloud delivery capability, which started as a direct response to the need to make Darwin financially self-sustaining. So it now has some commercial customers, still largely drawn from the wider community of Cambridge area hi-tech industries, though it is now open to all-comers.
One of its high profile commercial customers is the Lotus Formula 1 team, which is using Darwin to conduct Computational Fluid Dynamic analysis on its car’s aerodynamics.
The value of the SaaS approach for the research groups is an archetype of how SaaS can benefit many businesses. Their economic model is based on grants, which include an element ear-marked for purchasing the necessary IT resources. This means every research project has its own IT resources, all of which are different and in some way incompatible, if only because of differing project requirements.
The groups therefore end up with systems that cannot be integrated together and interoperated, so even though they have sufficient theoretical IT power for a new project, they cannot utilise it. So they buy more equipment.
With the introduction of the SaaS delivery of Darwin’s resources, they are now able to retire those old systems.
As well as using SaaS to widen its revenue generating possibilities, the Centre is also aiming for a wider payback for the community in general. By being broadly based on commodity hardware from a single supplier, Dell, it is able to refine its management and optimisation techniques and return that information to the community in the form of whitepapers and technical documentation on the techniques adopted. These will be open and freely available.
According to the Centre’s Director of HPC Services, Paul Calleja, the aim is to “can the experience required to set up and optimise the servers and systems needed to create an HPC service. A competent engineer will be able to obtain the specified systems, read the documents and quickly get an HPC system operational.”
An example of this is one of the first deliverables the Centre will be promoting – the Lustre Brick. This is a storage sub-system for the Lustre storage technology commonly used in HPC systems, and consists of a hardware specification of Dell devices, together with what Calleja calls the `cookbook’ of how to prepare and optimise the system.
Dell will itself be promoting an utilising the Cambridge whitepapers in helping to sell its own HPC systems to other users, which constitutes a useful quid pro quo for the company. And as HPC technologies are now moving into the upper reaches of mainstream IT for business and commerce, the potential market could be significant.
Calleja is also aware that the growing army of cloud service providers are a possible long-term market for the Centre’s services. They do not constitute part of any formal plan just yet, but he is aware that, as the overall market for HPC widens into more commercial areas, so the ability for service providers to partner with the Centre in order to aggregate its services into their own resources becomes a revenue opportunity.
This will also interest Dell, of course, for its experience with Darwin should give it some leverage in supplying service providers with suitable resources to meet wider cloud-based HPC market demand. “Commercially, the goal is a scaled-down, half-height rack that can be wheeled in to users’ locations and produce a reliable, repeatable system implementation,” said James Quarles, Dell’s Director of Public Sector Marketing.
Another deliverable that Calleja sees as both important and demonstrating how SaaS can benefit the HPC community, is the Remote Commodity Visualisation tool developed at the University. This can provide 2D and 3D visualisations of highly complex datasets to remote users that can be viewed on ordinary PCs and laptops. Not only is all the computational work performed on Darwin, but so is the rendering of the visualisation itself. So all that is sent to the user is a representation of what is being visualised, saving time, system cost and bandwidth.
[1]