Put simply, grid computing is a way of connecting a bunch of different computers so that unused processing power on those machines isn’t wasted but is instead put toward computing activity.
Grid computing prevents companies from having to shell out money on new hardware when chances are good they’ve already got what they need. It thus enables companies to get more out of their existing hardware investments. Indeed, early grid adopters such as Acxiom, Credit Suisse First Boston and Wachovia have realized such enviable performance increases and cost decreases through grid computing that I don’t understand why more CIOs aren’t pursuing it. I mean, come on, what company doesn’t want to show Wall Street the great return it’s achieving on its hardware assets?
Still, I have a few ideas why CIOs (especially those who don’t work at life sciences, automotive, animation and financial services companies) aren’t more interested in grid.
The first reason I see for grid avoidance is the perception that grid computing only makes sense for companies doing massive computations, as in industrial design, animation, financial risk analysis or mapping the human genome. This perception is well grounded as early adopters of grid computing have been in those industries.
The reason why grid has grown in those industries leads to the second reason why more CIOs aren’t interested in grid, which is the issue of scale, according to Jonathan Eunice, president of Illuminata, an IT research firm based in Nashua, N.H. Genomics companies, aircraft manufacturers and car markers need massive amounts of computing power to do their rendering and computer assisted design. Not everyone else does, says Eunice. “Not all applications need to take advantage of the scale that grid offers. If you’re running SAP R3, most SAP R3 databases are not scaled across 1,000 nodes.” He adds, “There are only so many companies that can afford thousand-node computers.”
Then again, a grid computing environment doesn’t need to have 1,000 or more processors to be called a grid or even to yield performance improvements. Bowne & Co., which prints companies’ SEC filings, built a computing grid as a proof of concept in 2003 that consisted of just two servers. Even so, Bowne reduced the time it takes to process data by 50 per cent, according to Ellen Kraus, Bowne’s chief IT architect. Bowne’s proof of concept shows that companies don’t need a grid with a thousand nodes for it to have a significant impact on business. Results like this again leave me wondering why more CIOs are lukewarm about stories on grid computing.
Well, maybe the explanation is not lack of need, but lack of ability…. The vast majority of companies’ business applications, especially the off-the-shelf packages, can’t run in a grid environment out of the box. They’re built to be installed on a dedicated server, not to be parceled out among, say, 20 different pieces of hardware. In fact, the way grid inherently works—by distributing processing across a variety of computers that don’t even all have to be running the same operating system—poses a fundamental threat to software companies’ traditional server-based licensing models.
In a grid environment, customers would need to be charged according to the amount of time an application runs on a particular server rather than the way software companies charge customers today, which is according to the number of computers an application runs on.
Since that could be 800 machines in a grid environment, software companies’ traditional licensing models completely negate the economies of scale that companies could get from grid, which is after all the main reason why companies build grid systems in the first place! The 451 Group, a technology research and analysis firm based in New York City, published a 77-page report last month concluding that software companies need to develop licensing models for grid if it’s to grow beyond the early adopters.
Until that happens, companies that want to try running their business applications on their own grids will have to write their own applications or use open source to get around the licensing problem, according to Illuminata’s Eunice. But that option raises another barrier to adoption: Most IT shops don’t have the resources to do all that software development in house.