Fermilab explores scientific use of cloud computing through FermiCloud
||Fermilab’s Computing Sector is investigating the possibilities of using virtualization and cloud computing at the laboratory. Photo: Hindrik S|
Ever on the lookout for how they can make scientists' lives easier, members of the Scientific Computing Division continually explore ways to provide researchers with more effective access to the laboratory's computer systems.
One of these undertakings is a project called FermiCloud, which aims to develop easy-to-use computing in two different but related areas: cloud services and virtualized machines.
"When we first proposed it in 2009, it was a bit of a radical idea," said Keith Chadwick, head of the Grid and Cloud Computing Department. "Now the user community is becoming more and more comfortable with it." Fermilab is investigating a number of virtualization services, including FermiCloud, for use at the laboratory.
A virtualized machine mimics the workings of a physical computer. It is built not from circuits and chips but from software and applications (which in turn run on hardware). One physical computer can run multiple virtual machines, each of which is tailored for a specific science investigation, allowing extremely effective use of available computing resources.
"You can support 10 different developers with different environments on one box," said Ruth Pordes, associate head of the Computing Division for Grids and Outreach.
In addition to providing virtualized machines, FermiCloud gives scientists a way to interact through a cloud interface.
Cloud services are distributed over a network, rather than located on an isolated machine.
Conveniently for the user, the responsibility for maintaining all the nuts and bolts in the cloud rests with the provider. The user can simply plug into the cloud and get to work.
Users could lease cloud services, such as Amazon's EC2 cloud. One experiment at KEK in Japan has considered EC2, and Brookhaven National Laboratory used EC2 from 2008 to 2009 to perform software analyses. The main drawback with commercial clouds, however, is the high cost of moving data in and out of them—and high-energy physicists have plenty of data to move.
FermiCloud is a private cloud and allows users to make full use of the existing Fermilab high-speed network infrastructure.
"It's already connected to the big data stores that we have, so we don't have to pay the data movement costs that you'd have to with a commercial cloud," Chadwick said.
Quick turnaround is another convenience. Rather than having to wait for weeks for new physical computers, users can be allocated computing on demand. This on-demand computer can be reconfigured for the next user when the previous user is finished.
A cloud often grants users access to external clouds through a process called cloudbursting. The Korean Institute for Science and Technology Information, for example, plans to combine its resources with those in FermiCloud, extending one cloud into another.
"It's very easy with FermiCloud to spin up a virtual machine, do a software test, gather the results, scratch your head a little bit to understand them and then repeat that process." Chadwick said. "It's all to help the scientists get the science done."