July 2008, an upgrade of University of Westminsterâ€™s previous 32 node High Performance Compute Cluster (HPCC) with a new 96 node HPCC, is providing a significant threefold increase in its contribution to the UKâ€™s grid computing facility for researchers, the National Grid Service â€“ NGS.
Established in 2004 and now with over 500 accredited users, the NGS is in use extensively across a wide range of research disciplines such as bioinformatics.
Amongst other projects, researchers from around the UK are accessing the NGS and drawing on its new additional power to more quickly understand how molecules, such as cancer or HIV interact with each other under certain stimuli. The faster researchers can do this, the more possibilities they can evaluate and the quicker potential life saving treatments can be discovered.
The new HPCC solutionâ€™s entire design, install and maintenance is provided by OCF, the UKâ€™s premier High Performance Computing integrator.
University of Westminster, the first London based university to be an accredited Partner of the NGS, also provides one of only two fully approved web-based portal interfaces to the NGS. This makes the grid accessible to all accredited users anywhere in the UK and, as a commonly used and understood interface, makes the grid very accessible to non-computing HPC users, such as biologists.
Professor Stephen Winter, Dean of School, School of Informatics, The University of Westminster comments: â€œThe previous 32-node cluster was considered a very useful contribution to the NGS and that has of course now been significantly enhanced by our 92-node upgrade.
Grid computing can bring a number of improvements both in terms of service delivery and performance. Simply creating a grid of desktops, and implementing your application in a grid solution might reduce the batch times, the time needed for your scientific research or trading algorithm from days to hours. This article from my former university discusses how they have leveraged grid or HPC technology to aid their research workload.
Implementing grid desktop can be a very efficient way of providing a grid infrastructure, key things to consider:
- A unified desktop grid in which all desktops on the network are ‘grid engines’, or a specific set of workstations or pcs which we use in a grid solution
- What management infrastructure do we need – a DataSynapse or Platform management server?
- Operation times – what hours can I safely use the desktop environment for my grid/hpc solution – 7pm to 7am?
- What layered components will each desktop need – do they need a database client loaded/configured
- The level of availability of each node – how many desktops do I expect in the grid?
- What management actions do we need to perform to have the desktops ready before or after grid – for example if we’re using the standard desktops in a grid application overnight, do we want them to reboot at 6am? Clear down or archive log files to a central location?
- What is the SLA of the desktop grid? The same as the server one?
- If we’re using shared desktop infrastructure what application(s) have priority – do the physics guys have priority over the biology guys?