Can high-performance computing, pioneered through NCSA's TeraGrid project and similar research initiatives, be harnessed for enterprise use?

---

by Edmund X. DeJesus, technical writer. Intel Corp.

Can high-performance computing, pioneered through NCSA's TeraGrid project and similar research initiatives, be harnessed for enterprise use? Our contributing author Ed DeJesus may convince you it can in the sections that follow. Drawing on experiences and findings of the TeraGrid initiative and others, he points to Explicitly Parallel Instruction Computing (EPIC), optimized use of addressable memory, and applications developed with operating system in mind as keys to Enterprise performance excellence.

Stunning scientific computing performance
Data mining, data visualization and other processor-intensive tasks put raw number-crunching power at the heart of enterprise computing. Development for processing power is something that researchers at the National Center for Supercomputing Applications (NCSA) working on the TeraGrid project know a little something about. As planned, TeraGrid will include a 0.4-teraflop cluster of over 3,000 Itanium processors, analyzing data, simulating a variety of real-world situations, and helping to solve complex scientific problems.

Editor's Tip: For an in-depth look at super computing on the TeraGrid project, check out Andrew Binstock's article, "Grid Supercomputer Demonstrates Itanium 2 Prowess." See Related Links at the end of this article."

First operational tests last year with a piece of the grid—the Titan cluster comprised of 160 dual-processor Itanium®-based computers—yielded eye-popping results. According to Karen Green, NCSA information officer, early runs with 256 processors produced results "more than two times faster than the previous best results," demonstrating that the cluster "scales extremely well up to 300 processors" and can "dramatically decrease the time needed to compute complicated molecular dynamics simulations."

Successful test results were matched by equally impressive outcomes when the Titan cluster was brought online in mid-April of this year. According to the Top 500 Super computer list, Titan has reached performance levels of 678 gigaflops (billions of calculations per second), with a theoretical maximum of 1 teraflop.

EPIC Plus addressable memory equals enterprise punch
Processor speed is one thing—and with the Itanium processor speed is significant—but the real story here is addressable memory.

"Clearly, one of the main advantages in using Itanium processors is the improved number crunching power of the EPIC architecture," explains Mike Showerman, acting technical program manager for the NCSA at the University of Illinois at Urbana-Champaign. At this level of computation, the size of addressable memory available to feed the processor becomes important. The microprocessor's 64-bit addressing allows researchers to move data faster with inexpensive memory.

How does this research experience translate to enterprise computing? "The design obviously lends itself well to scientific computing, but I think the Itanium processor has a good future in the data center," advises Showerman. "Instruction-level parallelism should provide great improvements for databases and data mining, and any data mining environment could benefit from additional memory." The combination of a larger memory address space with improved cache and memory performance can translate into significant improvements in the data center.

Lessons learned for commercial application
The research and pioneering efforts of the TeraGrid project and others yield important lessons for commercial developers.