IBM’S p680 Unix Server Deflates Superdome

The fact that IBM's newest high-end server has leapfrogged the competition in terms of performance is not news. What is news is that it will leapfrog competing products yet to be released.

IBM's pSeries 680 server, announced today, delivers better performance with fewer processors than similarly positioned products from Sun and Hewlett-Packard. Perhaps even more important is IBM's timing; HP's latest high-end product, dubbed Superdome, won't be ready for delivery until next year.

"We've got the benchmarks, and we think these prove that the p680 will set new standards for Unix servers for the foreseeable future," says Scott Firth, director of eServer products at IBM. "We'll be delivering these machines within a month, while all we see of Superdome is a lot of talk."

The server, which had been known by its code name Turbo, runs on processors incorporating silicon on insulator (SOI) technology. This approach increases the performance of copper-based chips by up to 30 percent by reducing electrical leakage among the transistors that make up processors. This allows greater performance from a smaller number of processors.

"This is a really big bump in terms of performance--roughly 50 percent above the best of the current products," says Richard Fichera, vice president and senior research fellow at Giga Information Group. "It's a significant new technology, and it definitely puts a lot of competitive pressure on HP and Sun."

Fewer processors could provide an extra dividend to users in lower costs for database and other per-processor software license fees, Fichera says.

The new server also adds a capacity-on-demand feature in the form of additional CPUs that are inactive--and free--until the customer requires them. Although HP announced this feature in its servers late last year, Firth says the concept came to the p680 as the result of healthy collaboration within IBM, which has resulted in concepts from the mainframe community migrating to its server line.

The p680 also comes with a service processor, an embedded monitoring tool that monitors the system and takes preventative, or corrective, actions automatically to head off problems. The system has the ability to deallocate CPU resources if it detects an impending CPU failure.

The p680 can be clustered for greater fault tolerance using the Blue Hammer Unix clustering system, which allows up to 16 of the servers with as many as 364 processors to be viewed as a single image. This technology is a direct result of IBM's work with very large supercomputing projects, such as the ASCI White computer it delivered to the Lawrence Livermore Laboratories this past summer.

IBM is targeting the server at e-business customers "who want to get online fast and scale quickly," says Firth. "We envision it as the back end of the Internet, handling thousands of transactions an hour for e-commerce."

The p680's impact on the high-end Unix server market could be dramatic. "HP hasn't released any benchmarks, and they could surprise us," says Fichera. "We can certainly expect some processor tweaks from them that will help performance in 2001. But this takes away a lot of Superdome's thunder, and it puts pressure on Sun. It won't derail Solaris, because there are so many users already, but it certainly won't help it either."

The result of this will be a performance arms race that will ultimately benefit customers. "At the extreme high end, performance is an eternal requirement, and these companies are zeroed in on surpassing each other," says Fichera. "This is a good time to be buying high-end servers."