But Grid computing also makes sense for more down-to-earth applications. Intel and Sun Microsystems use Grid computing in chip design. Pratt and Whitney runs computer simulations of turbine engines using Platform Computing’s LSF software. Caprion Pharmaceuticals in Montreal uses Sun’s GridEngine to speed up the analysis of human biological samples. A four-CPU server takes 720 hours to analyze one sample, but the Grid software lets the company assign up to 76 processors to the job.
“Organizations using it are the same ones that 15 years ago would have been using Cray computers,” says Brent Sleeper, principal of The Stencil Group, a San Francisco-based consulting firm specializing in Web services and enterprise software. “By going for a highly distributed approach they are getting much the same power but at a lower cost.”
Improving Endurance
Corporations typically utilize a three-year PC refresh cycle. But for the last several years, processor speed has greatly outstripped the needs of most business applications. While a 2.2 GHz Pentium 4 may shave a few milliseconds off processing a letter, that doesn’t translate into bottom-line advantage.
As a result, last year Gartner, Inc., of Stamford, Conn., switched from recommending a three-year refresh cycle to a staggered refresh. Those needing additional computing power, such as engineers, high-end graphics or data mining specialists are upgraded more frequently, while most users receive a new machine every four years.
This strategy, however, is flawed unless new machines retain their original capabilities. On Windows operating systems, in particular, workstations and servers suffer steady performance degradation due to disk fragmentation, anywhere from 20% to 200%, according to tests conducted by software testing firm NSTL, based in Conshohocken, Pa. Unless defragged regularly, these machines become sluggish long before they are scheduled for upgrade.
“Some companies, unaware of the impact of fragmentation, are likely to resolve such a performance impact with more expensive acquisitions of higher-performance hardware,” says International Data Corp. analyst Steve Widen. “However, it is just a matter of time before fragmentation impacts the new machines because this process only temporarily masks the performance problem.”
In addition, the excessive disk I/O caused by fragmentation leads to premature hard drive failure, wiping out the anticipated hardware cost savings. It is therefore imperative to install a networkable disk defragmentation program in order to keep the machines running at their peak and extend their useful life.
Boeing’s Space and Communications Division, for example, uses Diskeeper by Executive Software of Burbank, Calif.. “We can see a vast increase in performance using this new tool, especially on cluster servers within our 8 TB storage area network and Citrix Metaframe,” said Boeing Systems Administrator Tim McGovern.
As a result of regularly defragmenting all machines, Boeing was able to achieve high performance out of older 450 MHz Dell OptiPlex machines, delaying the need to replace them with higher-powered boxes.
Staying Compact
Storage demands continue to explode with ever-increasing file sizes. Enterprises typically find their needs doubling every year. While disk capacities are also growing, some companies are reducing their server room footprint using “pizza box” and “blade” servers.
When the Air Force’s Center for Research Support (CERES) at Schreiver Air Force Base in Colorado needed to boost its satellite simulation capacity, for example, it didn’t want to build a larger facility to host additional servers. CERES selected nStor Technologies, Inc.’s (based in San Diego) NexStor 3150 servers, which cram up to eight 73 Gigabyte disks (584 GB total) into a single 3.5-inch-high unit. This means up to 9.1 Terabytes of storage fits into a single rack.
Compaq ProLiant BL blade servers are another alternative. Each server — including hard disk, CPU and memory — sits on a card, rather than in its own box. Twenty 30GB servers reside in a single 5.25-inch enclosure sharing power supply, fan and wiring. Up to 280 servers fit into a standard rack.
The blade server architecture not only saves money by eliminating shared components (e.g. one fan and power supply, not 20), it also cuts power requirements by 75%.
This story first appeared on internet.com’s Datamation.