Suppose you needed the processing power of several thousand CPUs but didn't have a couple of billion dollars handy to purchase the hardware?
It's a common problem at today's IT centers where shrinking budgets conflict with the need to map the human genome or interpret signals from the radiotelescopes at Arecibo or plot a 3-dimensional graph of a gravity field. The solution, of course, is not to buy more hardware but to borrow the power of the hardware you need from where it's not being used. That's the theory behind "grid computing."
To take this notion out of the ridiculous and bring it closer to reality, few computers are actually put to work 24 hours each day, 7 days each week. Imagine a grid, or network, of such computers tied together by, for example, the Internet, and through which tasks are distributed. Why distribute anything? Perhaps a computer in Pennsylvania is at capacity, all processed up and slowed to a crawl, but there's a computer in Oregon where the business has closed for the day and the computers are idle. You could shunt some of the tasks from Pennsylvania to Oregon, speeding up things in the process. In exchange, if Oregon needs more computing horsepower, it can access the systems at Pennsylvania or Kentucky, or Sri Lanka or any other computer on its grid where load levels are low.
What the grid concept creates are virtual supercomputers and it's been the playground of academic and scientific institutions thus far. However, all of the signs indicate that it may soon become big business. IBM, Sun, Microsoft, Hewlett-Packard, Compaq--the big boys of the server world--are all interested in creating a world-wide pool of computing capacity. In the enterprise, the concept works something like this: One segment of the grid fields orders from online customers while another handles inventory searches and another does credit card billing and yet another might coordinate shipping. Each of these tasks may be handled by several computers as defined by the workload and the customer doesn't sit in front of the screen tapping the desktop, waiting for the whole thing to complete. The transaction data is then sent back to their respective companies at the end of the day.
Beyond the obvious benefit of increased customer satisfaction and the probability of repeat business, when you spread the work out over the grid the computing requirements needed to handle just the exchange of information at day's end are tremendously diminished over what they'd be if one company was forced to foot the bill for the hardware needed to process all of the transactions on its own.
As you might imagine, there is the potential for a logistical nightmare here. At the very least there are availability issues, privacy concerns, and security questions all waiting for answers. The Globus Project-- a research and development project focused on enabling the application of Grid concepts to scientific and engineering computing--has been working on those issues since 1996 (the grid concept itself is mid '90s) and has developed the Globus Toolkit that includes "…software for security, information infrastructure, resource management, data management, communication, fault detection, and portability." It's serious stuff. Both Microsoft and IBM are corporate partners with Globus.
You'll be hearing more and more about grid computing in the months to come but don't lose perspective. Someone must buy the bigger hardware so you'll only need the smaller boxes. Also, sharing someone else's computing capacity isn't likely to be free. As much as you might enjoy deferring large hardware purchases, those who make them would probably like to amortize some of their expenses. It just won't cost you as much.
More importantly, however, grid computing isn't about instantaneous transmission of information among nodes on the grid. It's about remote processing and cumulative reporting. Your data will be out of your control for some period of time and subject to power blackouts, denial of service attacks, and the full gamut of glitches that are possible on the Internet. Such possibilities may not have swayed the academic communities but they will become issues as grids migrate over to business.
What's your take on grid computing? How might grids fit into your company's infrastructure? Share your thoughts in our Talkback forum, or send an e-mail to Bill.
