The close-knit COSMOS consortium encompasses the leading UK researchers in both the study of the cosmic microwave sky and theoretical models for the origin of our universe and its large-scale structure. It includes CMB physicists who are involved in key experiments such as BOOMERanG, MAXIMA, VSA and the ESA Planck Surveyor satellite, as well as the originators of many of the key theoretical ideas in cosmology, together with pioneers in computational methods. It has a proven track record in using high performance supercomputers to achieve its ambitious scientific goals, pushing forward the exciting confrontation between models of the early universe and observational data. The consortium is built on research excellence with investigators drawn from eight departments, six of which achieved HEFCE's highest 5* rating (including DAMTP, The Institute of Astronomy and the Cavendish Laboratory in Cambridge).
The programme of research being undertaken by the COSMOS consortium will advance our understanding of the origin and structure of our universe, primarily through the scientific exploitation of the cosmic microwave sky. Our interdisciplinary scientific goals fall into three broad categories:
A particularly important aspect of work on COSMOS relates to the ESA Planck Surveyor satellite and is provided here as a concrete example. The goal of the Planck mission is to map the cosmic microwave background radiation (CMB) with unprecedented temperature sensitivity, frequency coverage and angular resolution. Planck is scheduled to be launched by ESA in 2007. Both Cambridge and Imperial College have a major role in the design and implementation of the Planck analysis pipeline. Within the Planck High Frequency Instrument consortium, the Cambridge Planck Analysis Centre (CPAC) has responsibility for Planck Level 3 analysis, which consists of a formidable set of high-level data analysis tasks which include separating contaminating foregrounds (including point sources) from the primordial CMB anisotropies, extracting optimal estimates of the spherical harmonics of temperature and polarization anisotropies from calibrated time ordered data, and computing optimal estimates of the power spectra of the temperature and polarization anisotropies. The London Planck Analysis Centre at Imperial shares responsibility for Planck Level 2 analysis with the Paris Planck consortium site.
The processing of Planck data presents major computational challenges and the manipulation of petabyte datasets with mirrored databases at different consortium centres. The necessary development within Planck of grid database applications such as replica management will closely parallel effort by other grid consortia. Close collaborative interactions between Cambridge, London, Paris and Munich as well as other sites, will be a major determining factor in the success of the Planck project.
In order to be able to make world class contributions to the study of the CMB and in the other major COSMOS areas, UK scientists need transparent grid-enabled access to world class facilities from their remote consortium sites. The core computational engine for the consortium is the National Cosmology Supercomputer, COSMOS, housed in DAMTP, an SGI Origin3800 server (64 cpus) with Infinite Reality graphics and an integrated Beowulf cluster (also 64 cpus). SRIF funding from four UK institutions has already been obtained to upgrade this further in early 2003 to a configuration with a 128cpu Origin and a much larger Beowulf cluster (several hundred cpus). This integrated shared-memory server and distributed-memory cluster provides an ideal environment for the rapid development, migration and optimization of grid-based applications. The COSMOS system will already incorporate `layer 3' grid technologies such as a high performance CXFS file system shared between the Origin and the cluster. Remote collaborating departments and institutions are also funded to obtain compatible distributed compute and visualization resources. Given this existing hardware infrastructure and the very close collaborative links within the COSMOS consortium, it is clear that enormous benefit is to be gained from fully implementing grid technologies. With technical assistance from SGI, the consortium will establish the National Cosmology Grid (cosmogrid) immediately following the COSMOS upgrade due at the end of this year. Key areas in which we aim to prioritize grid enhancements are the efficient management of grid resources through remote job scheduling and control, grid protocols for data transfer and replica management, and remote and collaborative visualization of large cosmological data sets. Advances in implementing and developing grid technologies in these areas will have benefits beyond the COSMOS consortium, for projects associated with the Cambridge E-science center and other strategically located remote sites.
The projects we propose are a continuation of a longstanding collaboration with SGI since 1997 in both job scheduling and visualization applications, including the donation in 1998 of an Infinite Reality graphics console. The COSMOS supercomputer has been selected this year as a Global Lighthouse site, one of only ten sites world-wide chosen by SGI for close collaboration and high priority technical support. The recently initiated work on remote collaborative visualization with Virtual Director from the NCSA has already been demonstrated at Supercomputing 2001 and iGrid2002. We also anticipate a linkup between the Cambridge eScience Access Grid facility and its equivalent in Manchester for a further Virtual Director demonstrations.