I got into this industry back in 1979 when I bought an Apple II and was instantly hooked on computers. So I’ve bought Apple products and watched the company with great interest for nearly three decades. Like many who’d grown up with Apple I was frustrated watching the company lose it’s hold on the market. (Do you remember when 2/3 of all internet servers were Macs?) And I was thrilled when Steve returned to the helm and made Apple a household name with the iPod.

Like many other folks I’d come to the conclusion that kids’ love of the iPod combined with the movement away from packaged software to SaaS was breaking Microsoft’s 20 year strangle-hold on the desktop. Perhaps, though, Steve isn’t waiting.

Gary Morgenthaler
speculated on GigaOM yesterday that
“. . . the Spaces feature of the forthcoming Lepoard version of OS X ” . . . “could allow you to seamlessly switch between Mac OS/X and MS Windows with a single keystroke.”

This is a tall order, even for Apple. However, if Apple were to attempt this, the correct solution wouldn’t be to run an OS X instance and a Vista instance, each with multiple applications. This would be like having two computers with a single monitor. Rather, Apple could enable applications to be installed into virtual appliances the same way we at 3tera do in the data center. The difference, is that each virtual appliance would run a single application. As a result, you wouldn’t switch between OS/X and Windows with a keystroke, but rather as you move between applications.

This is a longshot, but speculation is a fun sport.

Don Becker, of Beowulf fame, mentioned in a SearchOpenSource.com interview last month that “You’re always chasing device support. You always need to add just that one last device driver. That will never go away, and imagining that virtualization will make it go away is a mistake.”

While it’s true that new devices will constantly hit the market and need driver support, the real question is at what level those devices are integrated with the system. Don’s statement leads me to believe he’s proposing that application level software be rebuilt to support new hardware and that makes me shutter because it’s a 25 year step backwards.

I know it shows my age, but I can remember the days of the PDP 11 and the System 36 when it was common to tightly integrate applications and hardware. For instance, I remember upgrading a DEC accounting system when they released the new RL05 disk drive. They’d literally built the disk seek times into to the application as fatal timeouts.

Sounds extreme? It’s not so long ago that we had to enter bad sector maps for disk drives on our PCs. Or, perhaps you remember having to switch extended memory modes to run different applications.

I have to respectfully disagree with Don. While new drivers will always be needed, it’s the job of OS developers to provide a level of abstraction that keeps application developers from having to know about them, and more importantly from having to rewrite code to support them.

In a Network World article yesterday about a GridWorld panel on the promise and obstacles related to grid computing Sun’s VP of Engineering, Jim Parkinson, was quoted as saying one of the principle hurdles to grid computing is “. . . the lack of software development tools for building grid-aware applications . . ”

I don’t know Jim, so I don’t mean this as a swipe at him personally, but I think Jim’s statement shows the real problem - an assumption that people need to change the way they write code.

There are billions of lines of code in use today. There are hundreds of thousands of software engineers writing and maintaining that code. Trying to rebuild that investment is like carving a statue out of the Rock of Gibralter with a spork.

There is another way Jim, and that’s to make the grid support the software base that already exists.

The beauty of the internet is software becomes so easy to use that you can access hundreds of small but useful services without thinking about them . . . until they’re gone.

One site I find very useful for tracking the progress of new buzz words (remember, I’m a marketing dweeb) was Eirik Solheim’s Trendmapper. The concept of Trendmapper is simple, every night it submits a search for keywords submitted by users, gathers the results from the major search engines and graphs the result over time. In addition to seeing how certain terms do, or more often don’t, catch on it was interesting to see what terms people searched for. You’ll find “3tera,” “AppLogic” (both with a couple of exclusions that were necessary when we were small) as well as my name in the main search list. There are even a couple of brand names we considered before settling on 3tera and AppLogic.

Unfortunately, at the end of July, Trendmapper stopped updating its graphs. Eirik’s latest post over the weekend says simply “I need to do some major changes.”

Trendmapper’s just the latest in a long list of services to experience growing pains when the load gets beyond a few users. You may discount the significance of losing this little app, or you might just chalk it up to bad programming. However, I think it’s part of a more important trend. The internet isn’t just a collection of applications, but an ecosystem. Trendmapper was an excellent example of how a simple application could take data and turn it into information; an ever more impressive and necessary feat as the volume of data we all deal with grows geometrically. Sure, Google could provide the same data, but first it’d have to be a google sized market and I’m afraid most applications will simply never meet that criteria.

Brian Berliner wrote a flattering entry about 3tera recently, comparing our capabilities with EC2 and postulating about running AppLogic in front of EC2.

I certainly appreciate the comments about our system, but I think Brian’s bought into a common misconception that you can layer provisioning and higher level management systems and get the result AppLogic offers. It’s such a popular idea, in fact, that as I wrote once before I know of almost a dozen companies that tried to build something similar to EC2 and failed.

Virtual machine managers have their place, and there’s quite a few companies building them. However, trying to manage a large application by managing virtual machines is like trying to drive a car by manually firing spark plugs, pulsing the brakes and setting fluid flow levels to the engine. It doesn’t scale.

AppLogic provides an unprecendet level of simplicity in deploying, scaling and replicating distributed applications precisely because we DON’T layer the management in that way. Instead, our distributed kernel resides on all servers throughout the grid, creating a “complete” virtual environment for software. This means applications require absolutely no modification to code or configuration no matter how many copies you run or where in the world you run them.

So, while I can understand Brian’s desire to see AppLogic with EC2’s resources behind it, the architectural changes required might eliminate the features that attracted him to AppLogic in the first place.

Anyone who’s worked in a startup knows there’s a period during development when you wonder to yourself whether you’ll actually be able to build the shiny new widget you set out to. Actually, I this emotion may be stronger in those of us who can’t actually execute the design, but need to keep pumping up customers and analysts that it IS coming.

I wrote earlier, though, about that magic moment when you realize you’ve actually done it and the system works. It’s never the first burp from the system, but a point when you know it’ll actually work in production. With AppLogic that moment came when I was able to launch 30 crm apps from my sofa one Saturday while watching cartoons with my son.

Once engineering proves it can deliver is when we get to find out if marketing did it’s job. Did we make the right feature tradeoffs? Did we pick the right target market? What about the pricing model? Collateral? PR? When creating not just a new catagory, but a whole new way of conducting business every decision is revisited again, and again, and again.

And that brings me to another stress buster moment in the life of the startup. It didn’t come with the first contracts or even with the first customer announcement. Instead, while preparing for a staff meeting recently I suddenly realized we’re tracking more than 2 active evaluations for every employee in the company. In fact, with a few additional evals starting over the next couple weeks we may get close to 3 evals per employee. That realization is the moment I knew we’d gauged the market correctly.

Even so, there’s no time to rest, of course. We have a lot of work ahead of us to make 3tera successful. Life at 3 evals per employee (can I coin the “epe” unit of measure?) isn’t fun and games; it’s long hours, countless conference calls, too many emails to count, the crackberry buzzing at all hours, more than a few frayed nerves, and even a few angry spouses. Still, for some of us, this is what we live for.

It’s September and the whether’s warming up again in California. My UPS system dutifully beeped several times today to let me know of power fluctuations,
the most severe of which dimmed the lights momentarily. Now, being a confirmed geek, as I turned the AC down tonight I found myself
wondering if we’d have a repeat of July’s outages at Google,
Yahoo, MySpace and Bloglines. So I started searching for postings to see what folks are doing to prepare. If you spend much time online (and you’re here, so you must) you already know internet titans Yahoo, Google and MSN are spending billions building huge new data centers. Great what about the rest of us?

Eric Lundquist at eWeek (http://www.eweek.com/article2/0,1759,1995925,00.asp) pushes the old stalwarts of backup and data recovery planning. Oh, and he also suggests you push your server vendor for hardware that let’s you vary the power consumption with the load on the server.

The common wisdom is still to use redundant data centers, but the expense of doing so is beyond the reach of most startups. In fact, even well established service operators have trouble maintaining redundant sites. Setting up a redundant site is expensive and requires a lot of detailed planning. Maintaining it and ensuring it’ll pick up the slack when the primary goes offline is one of the toughest jobs in IT. Oh yeah, and if you’re eco-minded, it increases the overall use of power to serve the same service.

Virtualization pundits sing the praises of server consolidation to increase utilization, and thereby reduce power consumption. It’s a start, and eco-friendly, but doesn’t service online during an outage.

Once again, I have to admit I’m biased, but this all seems like linear thining. The answer is much simpler; eliminate your dependence on hardware altogether. As the hosting industry transitions to utility computing, you’ll not only be able to run your applications consuming less power, but maintaining a redundant site will become simple and cost effective. In fact, we’re working with software developers and hosting providers who want to go beyond simple - and make it automatic.

Eric Lai was among a host of writers to pick up on a recent Sage Research report that showed less than stellar interest in virtualization. While he reported the research factually, I don’t think he made the critical connection. The reason the survey showed low interest is that users have figured out what vendors and the surveyors haven’t - virtualization by itself isn’t a solution to anything.

Don’t get me wrong, I’m a BIG believer in virtualization AND feel the folks at VMware deserve the payday EMC gave them. However, all the hoopla about virtualization got way out of hand over the past twelve months.

To see what I mean, consider one result in Sage’s report, “. . . 28 percent of respondents said they were interested in server virtualisation to support a shift to utility computing . . .” What? Exactly how does Sage consider that even possible. Clearly they were coached by the vendor community.

So, here’s my somewhat less than humble opinion (because this time I’ve got more than a hundred customer meetings to back me up) virtualization, while useful, creates new problems. Virtual machine images are no easier to configure or manage than servers. In fact, they’re considerably harder to manage precisely because they’re virtual. To top it off, most tools being delivered to manage all those images are simply a rehash of server management; lists and databases. Therefore my conclusion is that interest in virtualization hasn’t waned at all, but that leading edge users have hit snags and are trying to figure out how to really make use of the technology. Of course, I have my own ideas about the solution