The concept of a grid is relatively simple: It generally refers to a form of distributed computing in which various technological components, such as PCs and storage devices, are linked across dispersed organizations and locations to solve a single large computational problem. As companies start to ponder the possibilities inherent in grid computing, the idea could morph into a grid that is self-contained within one company. However, the basic concept would remain the same, says Bret Greenstein, the project executive for grid computing at IBM in Armonk, N.Y. "We think that companies are going to leverage their distributed resources to get more value out of their infrastructure. Today, there are a lot of resources sitting idle -- companies use a distributed Unix server 10% to 20% of the time, on average, and with a PC, it's down to 1% to 2%."
And in these days of stringent fiscal frugality, any idea that promises to wring more worth out of existing resources is bound to get some attention.
Despite the allure of saving money, most companies haven't progressed very far along
"A lot of people are kicking the tires, but we're nowhere near beyond the early adopter stage," agrees Jamie Gruener, a senior analyst at the Yankee Group, a research company in Boston. He reports a lot of interest in the high performance computing space, healthcare, biotech, and the high end of electronic and automotive design. "There are some interesting places where computing power is finite and people need to do a lot of work," he says.
In fact, Claunch sees grid potential in such common business processes as financial and economic modeling, data mining and product design. For those CIOs actively engaged in tire-kicking for this new idea, our experts offer some procedural steps to success with grid computing.
Figure out what kind of grid you need.
Gruener says that there are two different kinds of grids, both of which answer different needs. CIOs need to analyze the problem they need to solve and choose their grid accordingly:
- A data grid is a grid used for sharing information. "It's more like accessing information over the Internet but with deeper content than you would traditionally get, or with more requirements for heavier lifting in terms of computational resources," says Gruener. For example, IBM and the University of Pennsylvania have teamed to build a grid that connects hospitals at the University of Pennsylvania, University of Chicago, University of North Carolina, and the Sunnybrook and Women's College Hospital in Toronto. It is funded by the National Library of Medicine. Researchers can use the grid to share information and analytical tools about mammograms.
- A computing grid, on the other hand, "is for the heavy crunching of numbers," says Gruener, and for telescoping the time necessary to arrive at the answer. The smallpox and Anthrax grids that IBM supports with grid.org are examples of that.
Expect custom software development work.
Grid computing is new enough that there are no technical standards or packaged software to put together a grid on the cheap. Greenstein says that companies need some sort of middleware software to tie together the computing elements on the grid, as well as grid software that knows what resources are available when and can schedule processing time accordingly. For example, IBM works with GloBIS, an open systems-based organization that has a grid toolkit.
In addition, companies will need to tweak any applications that run on the grid to accept distributed computational power. "Most people haven't built applications that can run one cycle at a time on different boxes," says Greenstein. On top of it all, expect some integration work in building a grid. "Most grids are now a somewhat customized mix of available products that allow you to roll your own software," Claunch says. "There's still a lot of custom development."
Think through the political ramifications
"There are a lot of governance-related issues to putting together a grid," Greenstein says. "Why should a particular department want to share its server pool?" Even though the resources belong to the company, the departmental view of personal ownership is hard to shake. Greenstein, who worked on an internal grid that runs at IBM, says that it takes some change management work to adjust the mindset of the company. However, companies that take the time to sit down and figure out some logical policies ahead of time shouldn't have a problem.Take, for example, the issue of funding. If a department pays for certain computing resources, and those resources are actually being used to crunch numbers via a grid by another department during idle cycles, who really should pay for the resources? Claunch recommends some sort of chargeback mechanism.
Security and privacy
Both issues have to be completely thought out by the grid masters, particularly if the grid will be a multicompany project. According to Claunch, companies that don't institute security measures run the risk of anybody who owns a machine on the grid being able to 'eavesdrop' on grid computations running on that unit. "If you're doing any sort of coopetition [such as joint oil exploration efforts, for example,] this would be a giant leap of faith," he says.
In spite of the fact that grid computing is clearly not yet ready for prime time, analysts remain bullish on its long-term prospects. For CIOs, this means that grid will start creeping into their strategy plans sooner rather than later. "If you're an aggressive technology adopter, you'll look at it today. If you're more conservative, it's something to monitor," says Gruener. Either way, it's clearly not something to ignore.
This was first published in April 2003