MIPs: The next commodity?

I'm not sure what to call myself though. In the tradition of ISPs (Internet Service Provider) and ASPs (Application Service Provider), should I call myself an IPP (an Internet Process Provider), a PSP (Processor Service Provider), or how about an IMP (an Internet MIPS Provider)?

Personally, I like IMP. One definition of the word "imp" is "small demon." That suits me since a lot of people think I'm a troublemaker. I like the MIPS part, too. MIPS stands for millions of instructions per second. It's a vendor-independent measure of processor capability. If I become an IMP, the IMPlication is that, when you come to me for some MIPS, you won't know or care what sort of processor or processors are in the background responding to your request. Actually, the way it probably would work is that some application or operating system of yours that was starving for more processing power would make that request on your behalf. It would all be transparent to you.

If technology is heading in that direction--and it appears to be doing so--we would essentially see what is the next step beyond Web services. Let's forget for a moment the nasty politics and patent mess that's holding back the evolution of Web services. The promise of Web services--and in particular the cross-platform nature of XML--is to achieve a degree of interoperability between applications regardless of the language in which they are written or the underlying operating system or platform on which they run. If the entire world moved to a Web services model tomorrow, the playing field between platforms such as Windows, Unix, Linux, and Java would be very much leveled.

For example, I could easily integrate a Web service-based spell checker into my e-mail client and all I would care about is that it checks my spelling. The provider of that Web service could just as easily be running that on Windows as it could on Java. My e-mail client could be based on Windows, Java, or Linux and it wouldn't matter to me. In fact, if that provider decided to switch platforms for whatever reasons, the migration would probably go undetected by me. As a layer of middleware that insulates the client from the server, Web services will have a commoditizing effect on operating systems.

Now, let's take that one step further. Suppose a similar layer of insulation existed between operating systems and processors. Operating systems wouldn't know or care where the MIPs were coming from, as long as they were coming. Take the idea yet another step: Let's assume the layer of insulation can also pool the MIPS to which it has access, regardless of the types of processors driving those MIPS. The MIPS could be provided by Intel Itanium, Sun SPARC, IBM Power, Compaq Alpha, or a mixture. Heck, why not toss in some old dinosaurs like a Motorola 6800?

A computing grid is similar to an electrical grid: You don't know how the electricity is generated --or from where. You only want to know that the power needed to run all your appliances is there when you need it. Hmmm… "App" could be short for appliance--or application. Take your pick.Will the processor turn virtual and will MIPS ever become a commodity like electricity? Probably. But it may take a while. Current prototypes suggest that the first commercial grids will have some dependencies. For example, there will be Linux grids that are only good for Linux-based applications and they may only be able to run applications designed for a specific processor architecture like IA-64 or IA-32.

But once those dependencies are gone, and much to the chagrin of processor manufacturers, IMPs will be a reality. In fact, in the same way that many local laws require utilities to buy back excess power from people and businesses that generate their own, the likelihood that we'll all be able to contribute excess MIPS to a grid is pretty high.

If the network becomes the computer, and grids end up leveling the playing field between processors, there's no question that we, as consumers of commoditized MIPS (whether we're consumers, businesses, or government/educational institutions), will end up paying dramatically less money for unprecedented access to computing power.

But what of the processor manufacturers? Considering that HP's PA-RISC and Compaq's Alpha's are being phased out, the only serious players left standing for server-based MIPS are Intel, Sun, and IBM. If you're an IMP, what would you pick to be the engine of your "processor farm" once you and all the other IMPs exhausted eBay of anything that resembled a MIPS for a next-to-nothing cost?

Grids are a love-hate affair for processor manufacturers. During our interview with Intel CEO Craig Barrett, he seemed genuinely excited about grids. But he shuddered at the idea of an insulation layer that virtualizes the processor and that takes Intel's proprietary instruction sets out of the game. If that happens, the door is opened for someone else like AMD, Transmeta, Sun, or IBM to turn up the heat. On the other hand, Intel has the sort of volume that could make its offerings the processor of choice for IMPs regardless of whether or not grids are vendor agnostic. That volume could easily mean that Intel will end up as the most cost-effective solution for delivering 32- or 64-bit MIPS.

The processor war is taking shape on many fronts. But, because of the innovation factor connected to grids, this battle will be the most fun to watch.

What do you think? Does the commoditization of MIPS sound like the promised land of processing power, or just a lot of hype? TalkBack to me.

Stay focused: Sign up for Tech Update Today, the daily e-mail newsletter for those who need to know.