Powers Unfiltered

As my loyal readers know, I recently spotted an article called “Linux is optimal OS for grid computing, says Oracle,” and I took a swipe at it because, well, I disagree. Today I’ll take a little more time to explain my comments about the Microsoft platform, and why we’ve bet our company on grid computing for Windows.

First, think for a minute about why grid computing is a good idea at all. Grid computing rides on top of some of the biggest trends in the IT industry:

• Networks get faster faster than CPUs get faster. Result: Every year, there’s a larger set of problems that can take advantage of distributed computing.
• 32-processor SMP machines are essentially the top of most product lines; these specialized machines are sold in such low volumes that their prices often exceed $1,000,000. Result: Raw processing power is vastly cheaper to buy in single- and dual-processor systems. These are produced in huge quantities, resulting in delightful economies of scale. Such boxes can be assembled into clusters, grids, or both.
• Building applications that take advantage of SMP boxes gets vastly harder the more CPUs you’re trying to manage, and there are very, very few applications that can take advantage of more than 8-way boxes. (John McCarthy famously summarized the situation: “Such systems tend to be immune to programming.”) Result: There are tremendous incentives to use a programming model that works well for many separate single- and dual-processor boxes.
• From Web services to enterprise SOA to a variety of Software as a Service models, more software is being exposed to more users as services every day. Result: Good services that deliver value (in the enterprise or beyond) soon become popular, and as a result consume higher and more variable amounts of computing resources. Even mildly compute-intensive services require a scalability strategy, and are a great fit for grid computing.
• Datacenters are filling up with ever-denser racks of blades and 1-U dual-processor boxes, which can be configured into clusters and/or grids. In many situations, rack space, floor space, power, cooling, and administration have all become more important limitations on scaling than hardware cost. Result: Grid computing technologies that can combine clusters with loosely-coupled resources outside the datacenter (including underutilized department servers and idle desktops) can greatly improve the economics in such situations.

These trends all cut in favor of grid computing — on any OS (or many OSes). Indeed, the above analysis says — “If it were easy, everybody would do it.”  Yet not everybody is doing it. Grid computing remains the “technology of the future,” as it has been for the past decade.  But with all these wonderful trends providing strong incentives toward grid computing, what’s been holding it back?

The answer is applications. Full stop.

It’s way too hard to adapt applications to most grid offerings.

And the applications are on Windows.  And the grid community has ignored mainstream Windows applications for too long.
Once again, as I brace for another episode of “Attack of the Rabid Penguins,” let’s get a few things clear.  Linux is a fine operating system.  For that matter, Mac OS X is a fine operating system, and there are plenty of others.  I’m not a “Microsoft suck-up,” and I don’t pull punches when Microsoft screws up. And I know that there are plenty of scientific and technical computing applications in academia, government, and enterprises that run on Linux, and OS X, and UNIX flavors of all sorts.

But I can count. When it comes to mainstream business applications, applications that run Web services, applications that run on the desktops and servers of CPU-hungry power users at businesses small, medium, and large — it’s a Windows world.

And when it comes to tools that let developers productively adapt those applications to the grid, it’s clearly a Microsoft world.  Microsoft has been about developers and developer productivity longer than it’s been about anything else, and that’s the Next Big Thing in the grid world.  (Anyone interested in recent figures on how Microsoft is doing with developers can check out this eWeek article, aptly titled “Microsoft: .NET Beat Java; Who’s Next?”.)
Find the applications.  Make it easy to adapt them to the grid.  Watch grid adoption soar.  Stay tuned as we make this happen!