Wednesday of this week Sun Microsystems CEO Jonathan Schwartz sent e-mail to the company explaining his latest brainstorm called Project Mercury. He wrote in part:

"...it's also my belief that we can't compete against our far larger competitors - in systems, storage, or software - without changing the rules of the game. Our volumes aren't as large, our relationships with some of the traditional IT suppliers aren't as leveraged, and our channels, brand and infrastructure don't yet afford us their reach or efficiency. I believe we have more innovation to offer, and our systems value proposition is vastly better, more by the day. But I also recognize we lose if we play by everyone else's rules.

"So this email is about changing the rules of the game, to put even more distance between us and our competition.

"It's about an SMI wide initiative called Project Mercury - which allows us to *lower* the price of our products to customers who agree to subscribe to Sun's service offerings. (Just like when you buy a cell phone in a shopping mall - sign up for service, you can even get the phone for free). Lower prices mean more demand, higher service attach means more margin and predictability (both prized by shareholders). At an SMI or systems level (vs. a component sale), Mercury improves our business model."

The entire message can be found in this week's links.

It is an interesting idea, essentially giving away hardware in exchange for signing long-term service agreements, trading hardware margins for service margins. I hope it works, really, but I doubt that it will. And the reason it is likely to fail is Schwartz's glib lack of understanding of his own people, who will tend to resist this change even if it means the death of their company.

What works so well for Gillette -- taking a loss on the razor to sell more blades -- may not work at Sun, where the sales department has always wanted to give away services in order to sell hardware. Now, Schwartz wants to give away hardware to sell services. He's going to have to change the company's reward system pretty radically to accomplish this, and Sun doesn't change that kind of thing easily, being something between a mob and a cult, rather than a company.

Can Sun make the necessary cultural change so that the whole company is heading in the same direction? Can they build and develop a good portfolio of products and services that will bring great value to their new business model? Most importantly, can Jonathan Schwartz get away from TELLING US that Sun is creating value and instead start a process where the results speak for themselves?

I don't think so. And what makes me believe this is the egocentric nature of the e-mail, itself. He never asks for cooperation from the company, never asks for help; he just says this is the way it is going to be. Maybe consensus is impossible to achieve at Sun, but not even bothering to seek it is a major mistake.

What Sun SHOULD do is partner with a PC maker to gain access to good low-cost hardware and recover those very scale effects that Schwartz acknowledges make Sun an underdog today. Sun could add real value and include good system management and protection tools to each PC. Between the desktop and traditional servers, Sun could create a family of turnkey appliances to provide file and print serving, email, etc. Anyone could do the above-the-PC stuff Schwartz is aiming for in his message. But no one is really looking at the total potential to deliver value to the customer. If Sun did, they could do IBM and HP a world of hurt without services even having to be involved.

Looking beyond Sun and its competitors, I am appalled at the lack of innovation taking place in the server business generally. With the cost of energy and data center space being what it is, making efficient use of servers is more important than ever. This is in large part driving a trend toward what's called "server virtualization" that comes down mainly to making four servers do the work of 10. Server virtualization relies on the fact that the smallest unit of computing power in many organizations is the single-processor server, yet many such servers aren't asked to do very much work at all. So why not consolidate a bunch of server applications in a single box that's larger than any of the servers it replaces but substantially smaller than their aggregate computing power? Server virtualization appeals to buyer and seller alike, so Microsoft and its few remaining competitors see a lot of potential in the technique. But what if you want to go the other direction and DISTRIBUTE service, rather than aggregating it? What if you want to go from one box to 10 or 100? What if, instead of making 10 servers run in four boxes you want to make 50 processors look like one? Now THAT's a lot harder, or has been until now.

Applications can be moved to faster servers, to multiprocessor servers, and ultimately to clusters of servers working together. But unless your application was designed to be distributed, the return in added application power from each added processor decreases until at eight processors or so it often isn't worth adding any more machines to the grid. Google can do it, sure. Google can cooperatively run tens of thousands of servers, dividing among them a single task, but what makes Google different from you or me is that company's 1000+ computer scientists who built the massively distributed system and keep it running.

Where's Googlization for the rest of us? It's called Appistry.

Appistry application fabrics take an application and spread it across tens, even hundreds of commodity processors. If you need more computing power, just add new nodes and the application will spread to them automatically, speeding up linearly as it does. If one or more nodes fail for any reason, they are simply ignored and the job continues minus those machines.

This is hot stuff and, talking with the St. Louis, Missouri-based company you'd think finding customers would be a breeze, but it wasn't at first. "We've always been about commoditization and virtualizing lots of machines to one, but it is hard to get people to save 80 percent on something that wasn't broken," said Appistry founder Bob Lozano, "Potential customers kept saying, 'Help me fix this problem I can't solve,' instead. Where customers were willing to try us was on these large computational or large data-intensive apps -- tasks for which there was no minimally acceptable solution."

So Appistry, which was invented with the idea of taking over mainframe-scale transaction processing (which it does very handily for customers like Sprint), finally found its niche doing tasks that simply couldn't be accomplished with mainframe technology.

One of the company's biggest customers is an unnamed intelligence agency. What does an intelligence agency have in common with Amazon.com? Both organizations need to analyze data quickly enough to take action on the basis of that data before the opportunity to take action is passed. For Amazon (not an Appistry customer, by the way) that means quickly displaying a list of possible additional purchases based on what you were just looking at and what you have bought in the past, and doing so before you have a chance to click away from the page. It's the same for the intelligence agency, being able to analyze terabytes of image and audio data and find an acceptable target before that target is out of range. This particular intelligence application that took five days to run on a single processor now runs in 48 minutes on a 150-processor application fabric. Apparently 48 minutes is fast enough.

"IT people with batch backgrounds have no sense of time," says Lozano. "They can gather data like crazy, but being able to analyze it in time to take action is something completely different. In batch mode you can come up with rules for optimizing transport routes for a shipper, for example, but with real time analysis you don't need rules, you just optimize the routes directly."

Installing an application fabric requires commodity PC hardware (anything from blades down to beige boxes and interconnected with the network of your choice), works with all major Intel-based operating systems, and can be accomplished on either a coarse-grained or fine-grained basis. The company's goal was always to take the fine-grained approach, which applies several layers of application fault tolerance and finely optimizes the system. But most customers are happy with the coarse-grained approach that simply spreads their .NET or Java or other application cooperatively across a fabric of servers by creating an XML process flow and a simple wrapper. This coarse-grained approach can ordinarily be up and running in one man-day according to the company. That's eight hours or less to make your application as many times faster as the number of processors you are willing to dedicate.

Appistry customers include GeoEye, the world's largest commercial satellite imaging company that uses an application fabric to process one terabyte of satellite image data every day and efficiently serve a market where per-image prices are dropping dramatically and customers like commercial fisherman want the answers to sophisticated questions such as "where are the fish?"

Another class of Appistry customers is Web 2.0 start-ups that can scale their application as their business grows, starting it on half a dozen processors and eventually building to hundreds if necessary by simply plugging them in. An example of this type of customer is DreamType, which uses an application fabric to do on-the-fly image customization for marketing materials. The demo application on their web site is running on that fabric.

But the greatest value of Appistry's approach to virtualization may come in the way it affects the work habits of users. Bioinformatics researchers at one pharmaceutical company were used to running an algorithm that took six hours to run on one processor, making it essentially a batch process. While the job was running the researcher could do little more than wait. Applied to an application fabric of 150 processors, that batch process now runs in 144 seconds, making it effectively interactive and changing completely the way the company uses its manpower and does research. THAT's innovation.

Now for some column housekeeping. Sometime in the next 30 days this column will morph overnight into a blog. I fought this for a long time but had to eventually bow to the inevitable. The attraction of a blog for readers will be the ease of commenting (that ease at present is, well, zero) and subscribing. So the whole exercise will become more interactive and I'll have to do more work.

The advantage for PBS is that all those blogospherical promotional tools can be applied to drag in more helpless readers and subject them to my radical ideas. I'll be waiting for that first comment from Jesse Helms.