SAN JOSE, Calif. — Corporate America is on the fast track to adopting clustering technology as a way to handle parallel processing, load balancing and fault tolerance, but most importantly, to save time and money.
Already the favored tool of scientists and universities, bringing the clustering issue to Main Street companies is being discussed in great detail at the inaugural ClusterWorld conference here this week.
For example Daimler Chrysler
and Shell Oil’s parent Royal Dutch Petroleum
say they are turning more and more to connecting banks of computers together in such a way that they behave like a single computer to solve complex problems.
But the biggest debate seems to be which vendor to go with when it comes to designing and configuring clustered systems.
The major pillars of IT like IBM
and Sun Microsystems
are pole positioning their strategies to help out. All four companies this week announced new hardware and/or services designed to help advance clustering, grid, distributed and High Performance Computing (HPC).
Among the many announcements here, Sun Wednesday said it is working with MSC.Software
to integrate Sun’s industry-leading Sun ONE Grid Engine software into its enterprise systems and high performance computing product portfolio. MSC.Software said it will market, implement and support the platform and corresponding services to help its manufacturing customers
Palo Alto, Calif.-based computer and printer maker HP this week said the National Center for Supercomputing Applications (NCSA) has selected its high-performance Linux cluster systems. The NCSA is using a 64-node cluster of Itanium 2-based HP rx2600 servers running Linux to test software, interconnects and compute-intensive scientific applications.
IBM Tuesday said it has completed and opened its first facility for delivering supercomputing power to customers over the Internet as part of its e-business on-demand computing strategy. The system currently consists of a cluster of IBM eServer xSeries Intel-based Linux systems and pSeries UNIX servers with disk storage.
Wednesday added a high-performance 64-bit system to its PowerEdge line, that the company said was designed for high-end computing tasks such as movie special effects and genomic research. The new systems will be available later this year in pre-tested configurations of 8-, 16-, 32-, 64- and 128-node clusters, running either 64-bit Red Hat Enterprise Linux or Windows 2003 Enterprise Server.
German Linux vendor SuSE said it has been tapped by supercomputing giant Cray
to drive key aspects of the U.S. Department of Energy’s new massively parallel processing (MPP) supercomputer called “Red Storm” at Sandia National Laboratories — which, when completed, will be the fastest super computer in the U.S. SuSE says the unit is expected to come online in the late 2004 timeframe and will use AMD’s
new 64-bit Opteron processors.
Clustering Comes Alive
By nature clustering is popular because it enables companies to use what they already have including their PCs and workstations. In addition, it’s relatively easy to add new CPUs simply by adding a new PC or server to the network.
For example, Daimler Chrysler Mainframes and High Performance Computing Manager John Picklo said his group uses a barebones three-person team to build systems for the carmaker’s engineers
“The main reason to do clustering is cost,” Picklo said during his keynote. “But if I can turn around a job in a day, that may make the difference in boosting a design of one of our models from a four-star to a five-star safety rating.”
Picklo said Daimler Chrysler currently uses HP Superdomes running Itanium, two 2 SGI
SMP servers to run the front end and an IBM 172-node Pentium cluster. He says that next generation speed technologies such as Gigbit Ethernet and Infiniband hold a lot of promise. The big challenge for Chrysler, Picklo says was integrating the systems together (he called it a bag of wires) and the software that runs it.
“There is a lot of cluster ware out there,” Picklo said. “Now I’ve got too many and its starting to impact my hardware decisions. The choice was either let the software influence my hardware decision or try to jumble it all together. Either way, I have my integrators run it the way I want before it hits the floor.
That’s where companies like Intel say they are working hard to make sure clustering and HPC move into the mainstream.
“Businesses like BP and Lockheed Martin are joining leading academic institutions and research labs in using powerful, standards-based Intel processors and software to tackle the world’s most complex computing challenges. With these powerful modeling and research systems, they can gain a tremendous business advantage,” said Intel Sales and Marketing vice president John Davies.