Big Blue launched its Deep Computing Capacity on Demand project in 2003 to let customers pay for short-term use of IBM's machines rather than buy their own. The program is designed to accommodate customers whose processing needs fluctuate, or that don't want to purchase and operate their own computing infrastructure.
The program has offered access to mainstream machines using Intel, Advanced Micro Devices or IBM processors running Linux or IBM's version of Unix. Now Blue Gene/L, a Linux system designed to speed life sciences research such as protein formation, is available, said David Gelardi, the vice president in charge of the program.
Blue Gene/L is an unusual design that packs 1,024 dual-core Power processors into a single rack. IBM in November began selling the systems for about $2 million per rack. A 16-rack model at Lawrence Livermore National Laboratory is currently the world's fastest machine and will be quadrupled in size by mid-2005.
In comparison, the on-demand program charges about $10,000 per week to use one-eighth of a Blue Gene/L rack, Gelardi said. At those rates, it makes financial sense to purchase a system outright for customers that plan to use a full system for more than six or nine months, he added.
IBM has only one rack available for outside use so far, but the company will install new models at its Rochester, Minn., facility as needed, Gelardi said.
IBM has put a top priority on high-performance technical computing, a market that includes supercomputing at its high-end extreme. The market is growing closer to mainstream computing as general-purpose systems get better at technical tasks and as ordinary customers perform technical work such as scouring customer data for purchasing patterns.
Petroleum Geo-Services and Halliburton subsidiary Landmark Graphics are among about 20 customers for IBM's general supercomputing-on-demand service. Hewlett-Packard also offers a processing-on-demand service, but the most aggressive advocate has been Sun Microsystems, which charges customers $1 per processor per hour to use its servers.
One customer for the Blue Gene/L system is QuantumBio, a start-up whose software uses physics simulations to predict the formation of complex molecules. The company sells its software to clients for tasks such as pharmaceutical drug discovery, but it now plans to offer the software as a service based on Blue Gene/L within the next month or so, said chief software engineer Lance Westerhoff.
Using the rival service from Sun is "something we'd consider," Westerhoff said, but the company right now is happy with IBM's customer support.
In addition to Blue Gene/L, IBM has more conventional products for the market, including clusters of lower-end and mid-range servers.
Even as IBM pushes into more exotic domains with its supercomputing designs, improvements in common technology are opening up the market dramatically. For example, Dell, which has little engineering expertise compared with IBM, Sun and HP, just sold a 210-server computing cluster to the University of California at San Diego.
