|
Grid Computing Takes Off in the Enterprise
Start thinking about computing power like you do about
electrical power: Pay only for how much you use.
by Mike Ellsworth
Right now, your computer and every other computer
in your enterprise is wasting resources. Millions of CPU cycles are doing
nothing more than warming the air. Megabytes of disk space sit unused. Your
network connection idles as you read downloaded Web pages. On a daily basis,
you and everyone you work with typically use only a small fraction of your
computers' capacity.
What if you could put all these underemployed resources to work? And what if you
could solve the big computing problems your organization has without spending
an extra dime on new hardware? That's the promise of grid computing, a method
of breaking a big computing problem into smaller chunks and farming them out to
dozens or hundreds of computers to work on.
Whether you steal idle cycles from your organization's desktop PCs or link
dedicated machines together to form a supercomputer, the result is a lot more
computer power to devote to massive computations like modeling weather systems,
oil deposits, or viruses; computing complex financial derivatives or insurance
risk portfolios; or simulating car crashes. And this computer power is provided
with a comparatively small incremental investmentusually miniscule when
compared to the cost of buying new hardware.
So what's not to love? You get more resources for your big problems, you get
better ROI on your computing investment, and you spend very little. That's why
more and more enterprises are looking into adopting grid computing to fill at
least some of their computing needs.
Grid is not a panacea, however, and the solutions available are not typically
going to help you with scaling Exchange servers or improving online transaction
speed, for example. Current grid solutions typically attack more strategic,
more massive computing problems, and are not yet suitable for some standard
enterprise applications, especially those with a low computing-to-I/O ratio.
Old Concept, New Implementation
Grid computing is not a new concept. In fact, researchers at Xerox's Palo Alto
Research Center first explored distributed computing in the 1970s. Along the
way, several high-profile grid projects have gained recognition, including the
rendering of the Toy Story movies through server farm and the
crackingthrough a grid network in 1997of Netscape's 56-bit encryption key.
One of the earliest prominent grid computing efforts was the SETI@Home project,
which links more than four million Internet-connected computers together into a
massive supercomputer. The Search for Extraterrestrial Intelligence project
sends each participating computer a chunk of radio telescope data to process.
When done, the computers send back the results and request more. Now four years
old, SETI@Home has 3.3 million users in 226 countries, has accumulated 1.4
million years of CPU time, is running across 136 operating system/hardware
combinations in 226 different countries, and has processed more than 800
million results. The capability of this grid averages more than 48 teraflops
per secondnot bad for an effort that has cost only a few hundred thousand
dollars.
Back to top
|