While writing material for 3tera’s website, I’ve been going back and re-reading a lot of source material, one piece
of which was Tim O’Reilly’s excelent article “What is Web 2.0“.

Sometimes while reading an article a second time, recent news and conversations lend new context and bring out new ideas, and that’s exactly what happened this time. One of the criteria Tim cited as differntiating Web 2.0 firms is a core competence in operations and that reminded me of an eWeek article about MySpace.

We all know web stalwarts like Yahoo, Google, Amazon, eBay and Salesforce are huge, of course. MySpace, however, is only three years old but they’re operating 3 data centers, 2,682 web servers, 90 Cache servers, 450 Dart Servers, 60 database servers, 150 media servers, and a 1,000 disk SAN. I’ve sold a LOT of enterprise gear over the years and only the largest of them add more than 1,000 servers per year.

Because the source of revenue for Web 2.0 firms is IT, they tend to have needs that correspond to traditional enterprises many times their size. The growth rate and the number of users that Web 2.0 companies support mean they’re encountering issues even the largest enterprises seldom see. To compound it all, their release schedule is often bi-weekly or faster.

And this is where I think Tim missed a point. In the same article he cites scripting languages like Perl as the tool of choice for building flexible operations. There’s a lot of Perl being used in just that way. The question is, should it? Scripts require just as much maintenance as any other lines of code. Properly written, they can provide a great deal of leverage for an operator. Poorly written, they become a nightmare of hidden dependencies. Either way, though, they’ve been a necesity, because there wasn’t a better way. Utility computing will be that better way. I’ve put gear been in the NYSE, the IRS, currency trading floors, banks, movie rendering studios, auto manufacturers, and even a couple in space. Still, Web 2.0 is the most demanding IT environment I’ve encountered. These operations run 24/7/365, with millions of users, volume doubles every quarter, and there are software releases every week.

All of this brought me to a conclusion I hadn’t expected. Web 2.0 isn’t different from enterprise IT; It’s out ahead. Web 2.0 is the new battlefront of IT.

An article on TechWorld a couple of days ago quoted Richard Warley, Savvis’ European MD, reiterating the well worn mantra that utility computing is “about being more flexible, and offering better value by arbitraging the unused computing cycles.” I’ll admit that when I first started thinking about utility computing the easily quantifiable goal of increasing CPU utilization was what came to my mind as well. However, I was wrong.

In early 2000 I was fortunate enough to have several customers buying networking gear from me that were early adopters of VMware for server consolidation. While they did save somewhat on hardware costs, none of them made the move to save money. In fact, most of them figured the savings were nill. Instead, security, backup and control were the reasons most often cited. Savings weren’t a driver because servers were already cheap, but managing the software that runs on them is costly. At the time server consolidation did nothing to help reduce the manpower associated with server management. VMware learned from their customers and over the last few years has geared much of the effort around reducing the time spent managing each image. For instance, making virtual machine images reusable so you can quickly restore a server by booting the VM images.

What I’m trying to get accross is that as we move toward utility computing, merely stuffing racks of servers into data centers is equivalent to server consolidation on a larger scale. The savings, if there are any, are incremental.

To realy make utility computing revolutionary requires moving beyond utilization as a metric. We need to deal with larger entities than server images. We need to move entire distributed applications onto the service, so we can minimize the manpower involved in deploying, managing and scaling applications. This is what 3tera has been working on for the past eighteen months.