You know there's been a big shift in the technical computing marketplace when Bill Gates headlines Supercomputing 2005, and talks about embracing the ever accelerating trend of low-cost, high-performance compute clusters. “Technical computing is crucial to the many discoveries that impact our quality of life,” said Gates. He was preaching to the converted, of course: technical users and vendors have known this all along. But although market growth has been propelled by UNIX® and Linux®-based servers (often run in clusters) from HP and others, the idea of mini-supercomputers running Microsoft® Windows® on people's desks isn't that far from reality—thanks to rapid advances in hardware and increasingly sophisticated software.
Technical computing capabilities more affordable
How quickly supercomputing zealots will embrace Windows over their beloved UNIX or Linux remains to be seen. The fact is, however, lines are blurring between business and scientific computing. “Industry-standard open source environments are making technical computing affordable for people who could never afford to buy these kinds of systems before,” observes Ed Turkel, manager of the product and technology marketing group for HP’s High-Performance Computing (HPC) Division. “Today, a two-person engineering shop can have its own cluster sitting in a corner. Or a research team can have their own system, rather than being dependent on some big shared system within an organization.” Indeed, the largest and fastest-growing areas of the HPC market are the low end—departmental and workgroup segments which include systems priced at less than $250,000. These make up more than half of the overall market, and according to IDC's initial estimates, are expected to grow at roughly 35 percent from 2004 to 2005, while the classic supercomputer "capability systems" market has reached a mature stage and is expected to see flat growth over the same period.
While this may come as news to many in the HPC segment, the shift hasn't happened overnight. For decades the industry has been moving away from monolithic multimillion-dollar proprietary supercomputers toward lower-cost configurations of VAX clusters in the mid-80s, then RISC/UNIX, and most recently, Linux and Windows on industry-standard processors from Intel and AMD.
Standards, improved storage and cooling technologies enhance clustering capabilities
Led by Digital Equipment Corp. in the mid-1980s, clustering is the biggest hardware trend in HPC today. The reason it's taken a while to gain acceptance is that not all supercomputer-based programs translate well onto the clustered model. “Some application vendors have to catch up,” says Turkel. For instance, in computer aided engineering (CAE), fluid dynamics applications run well on clusters but structural analysis applications tend not to. HP and others have been involved in government-funded research to create de facto standards such as Message Passing Interface which creates code to allow various end user applications to run in parallel on multiple machines. The HP-MPI implementation is helping independent software vendors (ISVs) to provide applications support for clusters, a major factor in clusters moving into the mainstream.
Vendors are also working to improve cluster storage performance as well as cooling technology (see Optimizing Data Centers for High-Density Computing), because when you put too many nodes in a rack, well, things can get hot. These days, it's not uncommon for a customer to install 1,000 nodes in one shot. “A 1,000-node cluster a few years ago was considered huge,” says Turkel.
“You couldn't do it without a major customer project. Now it's—I wouldn't say commonplace—but it's something we can do with standard configuration rules.” HP has built multiple 1,000-node and larger clusters, and is bidding on more all the time. “The size and scope are growing as the technology matures,” Turkel observes.
This up-scaling trend is important because traditional technical computing environments require ever-more processing power, in a more flexible package. A leading aircraft manufacturer that used to run large symmetrical multi-processor (SMP) systems for its structural analysis, fluid dynamics and impact analysis applications has been shifting to several clusters of 96 HP Integrity server nodes running HP-UX. “In the past they would have considered this stuff that the lunatic fringe (our euphemism for high-end) does,” says Turkel. “But we're taking this technology once thought of as bleeding-edge and making it industry-standard.”
To get started in a clustered environment, customers can opt for HP's Cluster Platform Express, aimed at the small to mid-sized cluster user from four to 32 nodes. Any user can go to the website and get an online quote for a cluster by entering key requirements (memory, processor, software, etc.). Price range: $40,000 to $500,000.
Information on HP’s High-Performance Computing Solutions
|
Verzija za ispis
