Case Studies > Non-profit

A BladeCenter Chemistry Set

Since its introduction in 2002, the IBM* BladeCenter* has made continued inroads within the IT marketplace, finding a welcome home in a variety of industries like retail, manufacturing and banking, all on its way to becoming one of the leading blade architectures available today.

The BladeCenter has also been embraced by academic institutions to serve a variety of tasks, both administrative and educational. Perhaps one of the best recent examples of the BladeCenter serving academia can be found in the hallowed halls of Harvard itself. Housed at 60 Oxford St. in Cambridge, Mass., resides a BladeCenter H chassis fully populated with LS-20 blades, a key to high-performance computing on which the Harvard chemistry department relies.

Computational Considerations

As you may suspect, in chemistry, the computational methods used vary quite a bit when it comes to their demands on IT resources. Such daunting terms as molecular orbital calculations and ab-initio methods imply very CPU-intensive calculations and can often use large amounts of storage, whereas molecular dynamics simulations can often push the limits of system memory.

"Many different aspects of cluster design can be stressed by different kinds of chemistry calculations," says Dr. Jerry Lotto, director for IT for Harvard's chemistry and chemical biology departments. "But even moderately complex systems push the limits of a desktop kind of an environment. If you want to do serious calculations, you really need fairly high-performance computing resources."

According to Lotto, the genesis of the BladeCenter cluster he eventually architected came about after thinking about how best to use advanced computational methods in the context of teaching. Harvard already had extensive research-oriented cluster resources being used, but students - and in some cases, educators - were loading them up with very complex - and even grand challenge - kinds of jobs. Newer students who were just starting out with simple projects were finding they often couldn't compete for the use of those existing computing resources.

"I wanted to build a modest-sized cluster that was - if not entirely dedicated - at least primarily-oriented to supporting teaching in chemistry," says Lotto. "We started looking into building a cluster that could have sufficient capacity to allow 20 to 50 students tackle a project simultaneously."

Next page: >>