Blades Spin ROI Potential

Computerworld - While total cost of ownership (TCO) and ROI issues are crucial to corporate IT managers, in the ivory tower world of supercomputing, the ongoing costs of operating these high-performance machines are only now starting to become important.

Until recently, most buyers of supercomputing power have paid little attention to the cost of power consumption, space and environmental requirements, says Wu Feng, technical staff member and team leader of research and development for advanced network technology at Los Alamos National Laboratory in New Mexico.

"In the next decade, size, power consumption, reliability and ease of administration will be the key performance issues in supercomputing. Bigger and faster machines simply won't be good enough," he explains.

Feng's team has been testing RLX Technologies Inc.'s RLX ServerBlades in floating-point applications since November and has found several advantages to using blade servers as commodity clustering servers for supercomputing. The blades from The Woodlands, Texas-based RLX were faster to deploy and are easier to manage than traditional server clusters.

"We were able to build our RLX cluster and get our code running in less than three hours, an effort that normally takes several days," says Feng.

Meanwhile, the space savings of the RLX System 324 "is a factor of eight times that of our traditional clusters," Feng notes. And the price/performance of the RLX ServerBlades is also impressive, he adds. "We calculated the [peak] price/performance ratio of a 24-blade configuration of the RLX ServerBlades to be $1.81 per millions of operations per second, vs. $6 to $9 per millions of operations per second for traditional supercomputers," Feng says.

Though first-generation blade servers are just starting to ship (many are still in beta testing), early adopters, including Los Alamos National Laboratory and Washington-based application service provider (ASP) Blackboard Inc., are finding that blade servers cost 30% to 50% less than traditional rack-mounted servers, with the biggest savings derived from their smaller size and low power-consumption costs.

Freed of the physical bulk and componentry of traditional servers, blades slide into slots on racks. In most cases, blade servers consist of processing and storage components housed in a rack unit that provides network and external storage connections, reducing both cabling and space requirements.

Analysts say cost savings increase the longer these systems are used. "Blade servers take up less space, generate less heat, use less power and don't need the environmental requirements of air conditioning or raised flooring, as larger servers require," says Tom Manter, research director at Aberdeen Group Inc. in Boston.

ROI Gotchas

The primary caveat in trying to achieve quick ROI on blade servers largely rests on how well the processing, networking and storage features are integrated. "Any cost savings can quickly be eaten away if maintaining blade servers becomes complex and time-consuming," says John Humphreys, an analyst at IDC in Framingham, Mass.