IBM, EMC, and Oracle Build 82-Terabyte Warehouse

Much of the world's attention is focused on the growing battle between Windows NT/2000, Linux, and Unix -- each competing for the hearts and minds of enterprise IT managers. Lately, however, IBM Corp. ( has been quietly promoting a hybrid operating environment that scales into multiterabyte range. Big Blue announced it built the world's largest data warehouse with its NUMA-Q technology, and is also releasing a database and the tools to support NUMA-Q operations. NUMA-Q, or Nonuniform Memory Access-Quad, was originally developed by Sequent Corp., which IBM acquired last year.

At its recent enterprise summit, IBM unveiled a version of its DB2 Universal Database configured to run on Intel processor-based NUMA-Q servers. The NUMA-Q server is scalable from four to 64 Intel Xeon processors, and includes a Fibre Channel I/O disk subsystem. In addition, the company unveiled a Web-based console that enables administrators to manage mixed Windows 2000 and Unix environments.

IBM is targeting these systems to the large commercial database market. The company recently touted what it calls the largest database in the world -- an 82-terabyte data warehouse, built on a NUMA-Q server with 48 Intel Xeon processors and running a 35 TB Oracle8i database (, built as a proof of concept for British Telecommunications PLC (BT, Data is stored on Symmetrix disk systems from EMC Corp. (

BT recently rolled out the first phase of the technology into production as a 13 TB data warehouse, designed to track its customer relationships within the United Kingdom. BT plans to consolidate large volumes of existing data currently held in disparate databases throughout the organization. "For BT to maintain the leading position in the market and improve relationships with customers, we need to intelligently use the large databases of information to develop customer focused activities," says Phil Dance, applications delivery manager at BT. He notes that "the proof of concept has reassured us that the data warehouse is flexible and robust enough to handle the job."

The 82 TB demonstration site shatters all previous records for database sizes, confirms Richard Winter, president of Winter Corp. (, which audited the tests. Winter notes that the BT proof-of-concept showed that the technology "scales beyond what was previously thought possible with a single database instance. Many people would have not expected that you could handle a database this large on a single-node system," he says.

In fact, this database is remarkably large for the NUMA architecture, Winter says. Ultimately, NUMA is better suited for large databases than SMP "because of NUMA's distributed memory," he explains. With NUMA, "each node -- or quad of four processors -- has memory physically dedicated to it. In SMP, all of the processors are sharing one memory. So memory bandwidth can become a limiting factor on SMP performance." The NUMA architecture also enables local processing, independent of the system bus, he adds.

If BT continues to build along the lines of the IBM model, it will be putting the world's largest data warehouse into production. The current size of BT's database -- 13 TB -- "represents 16 months of transactions," Winter explains. Currently, Wal-Mart has the largest data warehouse, which has a capacity between 15 and 50 TBs, he says.

Don't expect to see such large-scale databases running exclusively with Windows NT/2000 any time soon, Winter says. "The largest data warehouses are for the most part on Unix platforms or IBM mainframes. The largest databases today are going to be between 10 and 15 TBs of data. There's no NT system in production that I know of that's anywhere near that."

IBM's NUMA-Q architecture will support environments that include Windows NT/2000. The company recently unveiled a management console that provides a centralized, secure point of control for systems management across a mixed Windows NT/2000 application and Unix environment.

The console, called NUMACenter Director, is part of IBM's NUMACenter -- an application delivery engine that can support up to four IBM Netfinity application and Web servers running Windows NT/2000 and a NUMA-Q database server, with up to eight nodes housing up to 64 processors each running Unix in a single system environment with consolidated storage and back-up systems.

"NUMACenter brings a new level of management to the mixed environment that lowers the cost of administration and reduces complexity that can lead to unplanned down time," says Michael Kerr, vice president of product marketing for IBM's Web server division. "NUMACenter allows companies to add Windows NT/2000 resources in carefully planned increments to a reliable, scalable Unix environment, and manage both from a single point."