"Cosmology is a rapidly advancing and highly competitive field, driven by results," Hawking has said. "The COSMOS supercomputer has allowed us to turn ideas into concrete predictions, ensuring that UK cosmologists continue to make leading international contributions." Over time the project has evolved. "The questions we're trying to answer now are much more precisely defined," explains Dr Paul Shellard, the director of COSMOS.
"Ten years ago we were working with early universe theories, but cosmology has since become a much more quantitative science. We're having to make theoretical predictions of comparable quality and precision to the observations provided by the latest satellites and telescopes; and the observations themselves require much more computing power to analyse. That's why computers are so important. You can't analyse the observations without folding in your theory to see if it matches what you're seeing; and the observations themselves are huge datasets that need to be analysed to test the likelihood your theory is correct."
Sharing memory
To analyse and process the terabytes of data essential to the project, COSMOS employs an SGI Altix system running Linux. The system features cache-coherent global shared memory, integrated visualisation and hierarchical storage, and is heavily dependent on ongoing funding from the Research Council. "We are heavily over-subscribed at present, says Dr Paul Shellard, the director of COSMOS, "so we are looking forward to complementing our present Altix with cluster resources (which means upgrading to the latest ICE architecture). In the meantime, we are planning shortly to streamline the system infrastructure, and to give users more usable space, by increasing the storage capacity."
Shellard says that the system "makes available an entirely new threshold in shared-memory performance... and "will give us the ability to test our mathematical models of the universe seamlessly against the huge data sets being produced by cosmological observations."
"Cosmology moves forward through the ongoing confrontation between innovative theories of the early universe and an unprecedented flow of new experimental and observational data. The availability of a shared-memory programming model greatly helps us to get large-scale simulations up and running on short timescales. We have such a wide variety of users, both in subject area, code type and HPC experience that we need very flexible architecture that can respond positively to the wide variety of tasks we throw at it. We believe that the flexibility of these HPC resources makes an important contribution to UK leadership in cosmology internationally."
