When I think of grid computing, for some reason I think of my waffle iron. In addition to being grid-shaped, it makes high-performance waffles — nice and crisp with perfectly shaped ridges just waiting for the Vermont maple syrup layer. It’s also horizontally scalable — I can always get a friend to bring another waffle iron over to double my throughput.
But alas, neither my waffle iron nor grid computing seem to get much use. According to a recent report by the 451 Group, only a small number of today’s commercial applications that appear suitable for grids have actually been deployed on grids. The reason? ISV’s have other, higher priorities, according to report author and principal analyst William Fellows.
“Support for virtualization and service-oriented architecture is typically the key technical direction for ISVs that have not grid-enabled their software,” Fellows says. “ISVs often claim they’re still seeing limited demand for grid-enabled applications, although early adopters are often frustrated by the lack of movement from the software vendors,” he adds.
So maybe it’s a chicken-and-egg thing. Or maybe, like my waffle iron, grid computing is just too special-purpose to get the management mind share it needs to really catch on. I recently had lunch with a CIO who told me why he thought server virtualization is so hot right now: “We just want to be able to manage our distributed environment and the whole datacenter, like we manage the mainframe. We want it to be a no-brainer.”
So, is grid swimming against the tide of simplicity here? Kind of, according to the 451 Group. Grid-enablement means writing or rewriting apps and all the associated APIs, the report says. Challenges include job splitting, division and queuing, shared memory, and data management, as well as managing the potential for a “deadly embrace” or “deadlocking” between two apps sharing the same infrastructure. And, the report continues, “managing the output from a job run on a grid is another pain point, as are caching and the wider integration between storage, compute, and network resources.”
Still, as with my waffle iron, the desire to use the grid is there. Many smart minds are at work on software to better utilize datacenter assets and better manage workloads. Now if only they could invent a waffle iron that would also stir the batter.
Math geeks, part two: Last week I criticized a “nameless” research firm for playing fast and loose with its statistical methodology. I received several responses, seemingly all from Ph.D.’s, basically saying that I got it right but for the wrong reasons. What I’ve learned is that a small sample size is not necessarily a fatal flaw, but that failing to choose the right population rigorously to sample — the sampling frame — is where most research falls down. Failing to disclose the margin of error is adding insult to injury. Lesson learned: When in doubt, hire a Ph.D.
Math geeks, part three: While flipping through my copy of the Brown University Computer Science Department’s latest glossy magazine (lead article, “Randomness Is Beautiful, in Search of von Neumann”), I had a brief flashback to my late nights spent trying to get programs in Pascal to compile (sometimes did) and then run on an Apollo workstation (almost never could). To those about to code, I salute you!