Cambridge University To Offer Cloud-Based HPC

The University of Cambridge is planning a commercial high performance computing cloud, driven partly by a successful move to meter electricity usage in its departments.

Cambridge University’s high performance computing (HPC) service is planning to launch a commercial cloud. The service, currently offered to university departments, is upgrading with new servers, and planning to meet increased demand as the University applies new rules for electricity charging which are expected to drive University users to move computing tasks to a more efficient central resource.

cambridgedarwinhpc234.jpg

The public cloud is not yet formally announced, but will mainly serve local high-tech companies, said Cambridge University’s HPC director at the London launch of Intel’s Xeon 5000 “Nehalem” processor. Meanwhile, electricity savings will pay for an upgrade to Intel’s Nehalem servers, which will enable the service to meet increasing demand, partly driven by the University’s new electricity charging rules.

The University of Cambridge began charging its departments for the electricity they use this year – a move which could push more of them to use efficient centralised services. The University measures the energy used in each University building, and has set targets for energy use. In July 2009, departments that use more than their quota will be charged at current rates, while other will get a cash refund if they use less.

“The idea is to stimulate behaviour change by bringing the financial impact of energy usage closer to the end users, instead of being dealt with by the University’s central finances,” explained a spokesman for the University’s Energy Team. Previously Cambridge University shared energy costs based on the floorspace of each department, so less technical subjects like classics have been subsidising the more energy intensive technical subjects like engineering.

This visibility may increase demand on the university’s central computing resources, said Paul Calleja, director of the University of Cambridge’s high performance computing service, which runs the Darwin supercomputing cluster. As departments realise the actual cost of electricity used in computing, they may decide it is better to source it centrally and more efficiently. “Overall, departmental computing tends to be inefficient compared to a central service,” he said.

“I spend £360,000 a year on power,” said Calleja, who has always been charged for his department’s electricity use. “Last year, electricity in Cambridge went up by 57 percent.” To reduce power, and meet demands, Calleja is upgrading with Dell’s Nehalem-based Dell 11g servers. These apparently give a very quick return on investment, even when replacing two-year old machines based on Intel’s Xeon 5100 series processors, known as Woodcrest.

The commercial cloud, planned for this year, will run jobs on the Darwin supercomputer for high-tech companies around Cambridge, and should be able to easily beat other services such as Amazon and Google, on both price and performance, said Calleja: “Amazon and Google are really expensive,” he said, suggesting they are around four times the underlying cost price of computing on Darwin, “and they don’t offer high performance computing.”

As well as power, Calleja’s upgrades are driven by other physical factors. The HPC service is housed in the University’s New Museum Site – overlooking the shed where Crick and Watson discovered the structure of DNA – in an old building which can only hold a limited weight of servers and racks.
The old machines which are displaced are likely to go through brokers to the University’s departments, for training purposes, and for jobs which can’t be run centrally, said Calleja.

Overall, through turning out lights and not leaving devices on standby, the University hopes to cut its electricity usage by five percent this year – and many departments are on course to achieve this, according to the Energy Team.