Rich Wolski wrapping up his keynote address at eScience 2008.  Image courtesy of eScience 2008

Elastic Utility Computing Architecture for Linking Your Programs To Useful Systems

This mouthful-of-a-name came after the acronym, according to Rich Wolski, Professor in the Department of Computer Science at the University of California, Santa Barbara and lead developer of EUCALYPTUS (and coiner of the cleverly appropriate full name).

Wolski covered clouds, grids and the EUCALYPTUS architecture in his dynamic and forward-thinking keynote address last week at the eScience 2008 conference in Indianapolis.

Before starting EUCALYPTUS, he said, the team investigated both grids and clouds. These architectures share the metaphor in which users connect to an opaque resource that delivers computing and storage capability.  Grids use networks to link to distributed resources operated by independent institutions, and “glue” them together via middleware. Clouds are large, often clustered, infrastructures that typically depend on operating system virtualization (using software to act like a separate piece of hardware), and house processing power that users lease or rent for on-demand access.

Grids, Wolski’s team noted, require system administrators to reconfigure their systems and to install lots of software. Preferring to minimize the “violation of integrity” to systems (and to keep system admins as allies), Wolski said, when he turned to clouds he wasn’t yet sure what really made a “cloud” a “cloud”.  He also wasn’t sure if clouds could support scientific computing or data capture from sensors. In fact he’s still not sure.

Despite the unknowns, he and his team investigated Amazon and other big commercial computing operations based on cloud technology.  They developed EUCALYPTUS from open source web service technologies for cloud computing. It currently supports the Amazon EC2 and S3 infrastructure services.

How will it play with grids?

“The most important thing, I think, is for the grid community to experiment with the commercial clouds before making deployment or implementation decisions,” said Wolski. “There are certainly performance and scale question that need to be addressed, but as with the grid, some applications will work beautifully in the cloud and some will not.  It needs more study.”