IBM rolls out clustering for low, midrange users

IBM Corp. last week made a big push into the high-availability computing market with a new Unix server solution for small and midsize organizations and with new cluster management technology.

Clustering, which is designed for organizations that cannot afford to have their servers crash and disrupt operations, involves strapping together two server systems so that if one fails, the other can pick up the processing.

The new High Availability 50 (HA50) server brings together two IBM RISC System/6000 servers in a rack, with redundant critical components, including power supplies, adapters and fans.

IBM is targeting the new system, with a price beginning at $78,200, at commercial and government organizations that have shifted key applications to the Internet, said Rich Michos, director of clustered solutions for IBM's Server Group, Somers, N.Y. Agency World Wide Web servers and Internet-based procurement processing applications, by definition, need high availability, Michos said.

The new HA50 is largely a matter of packaging, testing and integration, Michos said. "And for the sum of the parts, instead of paying extra, you get a discount." As part of the development, IBM optimized the system to run such widely used applications as its own DB2 database, Lotus Development Corp.'s Domino and Oracle Corp.'s database management and financial software.

Michos described the HA50 as a very scaled-down version of the RS/6000 Scalable Parallel (SP), which is the company's high-end clustering solution designed to run very high-performance applications. The HA50 can run up to four reduced instruction-set computer processors, and up to eight HA50s can be clustered, Michos said.

It is a business, not a technology, announcement, said Harvey Hindin, director of the High Availability and Cluster Service for D.H. Brown & Associates, New York. The HA50 competes in price with a solution from Compaq Computer Corp. based on less expensive Microsoft Corp. Windows NT and Intel Corp. processors, he said, and Windows NT clustering is "two years behind."

IBM also announced that it is developing cluster management technology that will allow or-ganizations to attach an IBM Netfinity Windows NT server cluster to an RS/6000 SP configuration and manage both from a single control point.

"We feel, across the board, a growing demand for heterogeneous systems," said Bill Zeitler, general manager of server brand marketing with the IBM Server Group. IBM will begin shipping the technology later this year. About a year from now, the company will add high-speed connectivity to that high-availability solution, Zeitler said.

The management of heterogeneous operating systems from a single point would be unique and should reduce management costs dramatically, said David Floyer, director of research with International Data Corp., Framingham, Mass. For example, the cost of managing storage drops from 55 percent to 15 percent of the cost of hardware in a clustered topology, according to IDC.

Floyer predicts an "enormous takeoff" in clustered configuration sales as demand grows at the low end. By about 2002, Floyer said, 40 percent of all systems will be clustered,

driven primarily by administration and secondarily by high-availability requirements.

The market is "gigantic," said Ed Schaider, vice president of the Standish Group, Los Gatos, Calif. The Internet, in particular, drives the demand for availability and scalability of clustered systems.

Adams is a free-lance writer based in Alexandria, Va.