Universities across the UK are investing in increasingly powerful computing resources to support the work of their research scientists and in doing so they have to invest every three or four years in a new supercomputer. Conservative estimates from academics at the University of Westminster put potential savings for universities using the desktop Grid system at GBP 500,000 every four years, a significant amount for universities trying to minimise costs, particularly in the current environment.
The Higher Education Funding Council for England (HEFCE) recently announced a GBP 12.5 million programme to promote Cloud computing in universities. The Grid system implemented at the University of Westminster is an illustration of how universities can implement cost savings.
The easy to manage Grid system the University of Westminster has implemented maximises the essential resources a university already owns. While PCs are utilised during the teaching periods, there are large periods of time during the night or holidays when computer labs are idle. These unused PCs are employed to run computation-intensive tasks for university researchers. Bio-scientists use powerful simulation programmes, linking results together in a sophisticated work flow, to understand how molecules potentially bind together in a process known as docking. Researchers using the Westminster Local Desktop Grid have found that they can shorten a typical execution time from weeks to hours. This rapid acceleration of the scientific experimental method helps the scientists in their particular endeavours - to help bio-scientists design new vaccines more quickly for example, or environmental scientists to more rapidly identify environmental risks and damage.
Professor Stephen Winter from the Centre for Parallel Computing at the University of Westminster explained: "The desktop Grid system implemented at the University of Westminster consists of 1500 laboratory PCs - roughly half the total number of PCs in the university which is equivalent, in raw computational power terms, to a GBP 500,000 cluster procurement or supercomputer. But it has the added bonus that, unlike a cluster procurement, the desktop grid replacement costs are zero, since the PCs themselves are continuously replaced from existing budgets, while an installed cluster typically needs replacing on a three or four year cycle incurring all of the costs associated with this. Self-replenishment also means that the performance of a desktop Grid always improves over time: old PCs are always replaced with much better, higher performance ones, so the desktop Grid always keeps pace with PC performance improvements. A good example of this currently is the growing use of high powered graphic processors (GPUs) in standard PCs for general purpose computation. GPUs are very powerful computing devices in their own right, and are the basis of a new generation of computing clusters known 'GPU farms'. Being inside a PC, the onboard GPU's power can be easily and seamlessly harvested by the desktop Grid."
Professor Winter continued: "Many British universities may think that they are unable to afford a supercomputer but the fact is that they are sitting on resources that, utilised in the correct way, can be the equivalent to hundreds of thouands of pounds worth of computing power. Desktop Grids are also green, in the sense that they soak up the spare computing capacity of desktop PCs, focussing computations on fewer, and therefore more energy efficient, resources."
The Centre for Parallel Computing is a partner in two current EU Framework 7-funded Desktop Grid projects; EDGI: European Desktop Grid Initiative and DEGISCO: Desktop Grids for International Scientific Collaboration. The Centre was also a partner in a predecessor project EDGeS which completed in May 2010.