IBM upgrades SP server

IBM Corp. last week unleashed a more powerful version of its RS/6000 SP server system, which brings supercomputer power to bear on data mining, scientific modeling and simulation, and online transaction processing. The new RS/6000 SP is a faster version of the system that powered the IBMdesigned s

IBM Corp. last week unleashed a more powerful version of its RS/6000 SP server system, which brings supercomputer power to bear on data mining, scientific modeling and simulation, and online transaction processing.

The new RS/6000 SP is a faster version of the system that powered the IBM-designed supercomputer Deep Blue to a 1997 victory against chess virtuoso Garry Kasparov in a grueling six-game match. Providing that extra power and speed is IBM's new Power3 microprocessor, which is capable of churning through 2 billion instructions per second—more than twice as powerful as the Power2 Super Chip.

David Turek, director of technology strategy and business opportunities for IBM's server group, described the RS/6000 SP as several "individual and self-contained computers connected with a high-speed switch." In fact, each server is stacked in a rack-mount frame that is capable of holding up to 16 systems.

The RS/6000 SP is the central piece of an initiative IBM dubs Deep Computing, referring to the ability of the Deep Blue supercomputer to crunch through millions of scientific and mathematical calculations. The power behind IBM's Deep Computing initiative is the Power3 symmetrical multiprocessing (SMP) node architecture, which provides improved system bandwidth and performance by allowing various functions and calculations to be offloaded to individual nodes, or processors.

Each node contains at least one microprocessor and its own RAM. Different types of nodes can be used together to optimize performance on large computing tasks. For example, several nodes can be used for World Wide Web server operations, while others can be dedicated to scientific calculations.

According to Turek, a system like the RS/6000 SP is perfectly suited for customers interested in consolidating their servers to improve performance and reduce operating costs or to enhance their ability to handle large data mining requirements of high-volume online transaction processing.

One of the largest configurations of the system is at the Lawrence Livermore National Laboratory, where scientists use 1,463 nodes to do their data crunching, Turek said. Commercial firms such as the online brokerage firm Charles Schwab & Co. also use an RS/6000 SP—with more than 40 nodes—to handle the company's 2 million Internet trading customers.

The RS/6000 SP is "supercomputing at your fingertips," Turek said. "The National Weather Service sees it as a supercomputer, while Charles Schwab sees it as a scalable server."

Because the switches used to connect the systems into a coherent cluster are so fast, "distance is irrelevant in terms of gross bandwidth and system performance," Turek said.

In fact, although IBM has developed a switch that is capable of handling in excess of 32,000 nodes, the cost of acquiring and operating a system on that level is beyond what anybody would want to incur, he said. However, the system is capable of reaching bandwidth performance levels of 160 megabytes/sec, Turek said.

Brad Day, vice president and senior analyst with Giga Information Group Inc., said IBM's new RS/6000 SP announcement significantly boosts the SP's appeal from a performance and a scalability perspective. "Now the SP can be a switch-hitter" in terms of availability and performance, Day said.

This upgrade puts the SP in the class of significant performance and scalability and provides that mix of capabilities to companies that need a little of both, Day said.

Joyce Becknell, director of Unix at Aberdeen Group, a Boston-based technology research firm, said IBM's improvements to the RS/6000 SP have "raised the bar on the SP's performance and brought it to a new level." Although the announcement represents more of an improvement on the technical computing end of the spectrum than in the commercial, IBM's Deep Computing initiative will benefit companies operating in the intersection of these two areas, namely business intelligence and data warehousing, Becknell said.

To date, the National Weather Service has put the RS/6000 SP to use supporting a $35 million program to improve the ability to predict weather patterns by running highly sophisticated models and simulations. In addition, IBM will be installing a series of SP systems at the super-secret National Security Agency later this year, Turek said.

***

RS/6000 SP

* 200 MHz Power3 SMP microprocessor

* Up to 16 systems per rack

* High-speed switch capable of 160 megabytes/sec

* Switch supports up to 32,000 nodes

* Integrated Ultra SCSI and Ethernet (10/100 megabits/sec) controllers

* AIX Version 4.3.2 operating system