The Magellan management and network control racks at NERSC. To test cloud computing for scientific capability, NERSC and the ALCF installed purpose-built test beds for running scientific applications on the IBM iDataplex cluster. Magellan systems at both NERSC and the ALCF will be built using QDR InfiniBand fabric. Image courtesy R. Kaltschmidt, LBNL

It’s clear that cloud computing is today’s hot computing buzzword. All the cool kids are doing it, and scientists are no exception. But how much do we really know about the intersection between science and cloud?

“Cloud computing has become a very exciting new field with several companies making offerings that are already being used by scientists around the world,” said Pete Beckman, director of Argonne Leadership Computing Facility and leader of the ALCF Magellan team. “The question the Department of Energy has is pretty straightforward: what kind of science can be done on clouds, and are there specializations or customizations that we can do on the software to get more science out of clouds?”

Jeff Broughton, head of the systems department at the National Energy Research Scientific Computing Center and lead for the NERSC Magellan team, explained, “There are a couple of different aspects to that” question.

The first is whether or not cloud technology — as modeled by companies such as Amazon, Microsoft and Yahoo — is applicable to science. The second, Broughton continued, is whether or not cloud computing can be used to enable science to be done in a different way.

To answer those questions, the NERSC team decided to set up several small clouds with differing architectures.

“We’re going to be running applications in all of those environments to get performance information,” Broughton said. Then, they’ll try to understand whether or not there are specific disciplines of science, or specific kinds of algorithms that are well-suited to cloud computing.”

Preliminary results suggest that some application areas are not well-suited to cloud computing, while others, such as some aspects of genome processing, are good matches for the technology.

The main system console for Magellan at NERSC. Image courtesy R. Kaltschmidt, LBNL

The ALCF team has also made significant progress.

“We have a large system currently deployed and we’ve opened it up for a variety of users,” Beckman said. Already, about 20 projects have applied for time on the system and had their applications approved. These range from earth science to bioinformatics to computer science.

“The Open Science Grid has an allocation on our machine and will be exploring how to use cloud computing in the context of this experiment,” Beckman added.

Users on the ALCF system will also have the option of getting access to the actual compute node in order to install their own operating system. For example, the project doing genome work on the ALCF Magellan are installing whole software stacks capable of using their tool chain — a feature not supported by standard cluster software.

It’s an exciting feature, but it introduces some new problems.

“The notion of a cloud, and running your own software stack, poses new challenges that the software and science communities have not addressed before,” Beckman said. “As a simple example, if you want to show up and run your own operating system, what sort of vulnerability testing should it undergo? What kind of probes should it undergo to make sure it’s not accidentally opening things up and causing mischief?”

By the time the funding draws to a close in September 2011, the two sites are expected to demonstrate effective cross-site administration. Realizing that mandate means that each site will have to ensure interoperability from both a technological and policy viewpoint.

“The people already talk; we work together on the software and architecture all the time,” Beckman said. “The more difficult technical question is what sorts of software as a service or storage as a service might make sense to fluidly exchange and transfer between multiple sites.”