Why HPC Is Flying High

SANTA CLARA, Calif.—IT's high-performance computing sector compares nicely with NASA.

Many of the innovations of daily life we now take for granted—such as satellite television, Velcro, laser pointers and instant fruit drinks—originated or were commercialized through NASA and reside among NASA's 6,300 patents. Similarly, HPC is the innovation engine of enterprise IT.

High-performance disk drives, super-fast I/O channels, petabyte-capacity storage, deduplication, and virtualized servers and storage are just a few of the inventions that originated in HPC labs and are widely available now in less expensive systems.

Sun Microsystems CEO Jonathan Schwartz said recently that rapidly growing data loads generated within enterprises mean that forms of high-performance computing will be coming to every industry sector sooner or later.

That, he said, means that many businesses in numerous industries will be turning to companies like his to help update or replace their systems. Of course, he did not mention that these new customers will also be visiting IBM, Hewlett-Packard, Dell, EMC, Fujitsu, Hitachi, and other general IT infrastructure providers. 

Schwartz added that due to the increasing amount of data being saved by businesses due to recent federal regulations—both in the United States and in numerous other countries—data center storage systems must be faster and more efficient than ever, which requires the highest level of software and hardware performance.

As a result, companies like Sun Microsystems, Dell, HP, IBM, COPAN Systems and a number of startups have made a high priority of investing time and capital in storage system development.

Speed and storage demands are driving an HPC renaissance

The swift rise of cloud computing—applications made available as on-demand services for enterprises and consumers over the Web—is now requiring HPC and "super" storage at all levels, Platform Computing founder and CEO Songnian Zhou told an audience of several hundred IT managers and developers here at Platform Global Conference, held May 19-21.

Platform Computing makes specialized management software for HPC data centers serving sectors such as the financial market, earth science, oil and gas exploration, health care, and government and military installations.

"Current data centers, most of them built more than 10 years ago, are costly to run and not very efficient in using power resources," Zhou said. "What IT managers and CIOs need when they are looking to upgrade are agile, scalable, more powerful, more cost-effective servers and storage systems that use more automation, share resources, use less power and run on commodity hardware.

"Yet these new systems must be able to deliver powerful Web services 24/7. This is what HPC brings to the table."

IDC Vice President Earl C. Joseph, executive director of the HPC User Forum, told the conference in a keynote address that high-performance computing was an $11.5 billion business in 2007 and is growing at a 19.5 percent clip per year.

"This makes HPC by far the fastest-growing sector of IT worldwide," Joseph said. "In fact, we at IDC this week are announcing an entire new sector of research: high-performance computing management software."

Mainframe resurgent, but x86 remains dominant

x86-based server clusters are now making up more than two-thirds of the data center hardware market, Joseph said, and sales are continuing to grow at 60 percent annual rate. Hewlett-Packard's C-Series and blades from Sun, IBM, Dell and Hitachi are leading the pack.

"Only five years ago," Joseph said, "x86-type servers only occupied a small sliver of the market. The growth there has been nothing short of phenomenal."

Blade servers are powering about 40 percent of all new HPC clustered systems, Joseph said. "Although x86-powered blades are dominating the data center space, new mainframes—with all their value-added features and improvement—are starting to make a comeback," he said.

Linux is steadily replacing Unix as the HPC operating system of choice, Joseph said. "Unix systems just aren't being replaced by Unix anymore," he said.

Thanks to the new, less expensive servers, the prices per processor have dropped substantially, Joseph said.

"Generally, it costs about $2,000 per x86 processor, while RISC-based processors are going for about $7,800," he said. "Itaniums cost about $9,000 each; Vectors are by far the most expensive at about $54,000."

There is quite a horse race among the leading HPC vendors, Joseph said. HP and IBM are leading the market at about 32 percent market share apiece, followed by Dell (18 percent), Sun (5 percent) and SGI (2 percent).

"Only a few years ago, that lineup was very different," Joseph said, adding that the rise of commodity-type hardware has indeed changed the face of computing—even at the highest levels of HPC.

"Don't worry, there's still a lot of good work going on in mainframes, some of which have been around a long time," he said. "But HPC clusters are where the trend is and will continue to be for several years, we believe."