The move to utility computing may be hedged around with discussions about outsourcing, and hardware platforms, but actually it is much simpler than that, according to Marc Andreessen, chairman of Opsware. In fact, it is just one more aspect of a long march from client-server computing to web-based systems, a march that will rewrite the business models of server and system management companies.

It’s not surprising to hear the erstwhile uber-browser maker still puts the Web front and centre. But with his new project, DCML, an effort to create a standard for data centre integration, he has something that should finally stop journalists asking about his days in the browser wars between Netscape and Microsoft six years ago.

For one of the industry’s fastest achievers, he takes a surprisingly long view of the changing tech landscape. “The biggest thing in the industry is the transition from client-server to the Web,” he says, a cycle he expects to take twenty-five years and is only reaching critical mass now.

We might get depressed about the decline in technology spending, he says, but web applications are being developed,and the majority of the investment is being lost from the old guard. “Client server companies like Sun and Oracle are doing poorly, “ he says. “The big Unix businesses are in decline and the next five years will see a huge move to Intel and Linux.”

If anything the poor financial climate is forcing the issue of utility computing that DCML addresses, he says: “You have 5000 to 20,000 servers running thousands of applications. It is critical for the business and you have to run the system without 2000 people.”

The Opsware story
Opsware is the new name for Loudcloud, the Web-based outsourcing company that Andreessen founded in 1999, some time after AOL bought his first company, Netscape. Loudcloud was one of a large number of application service provider (ASP) start-ups that came and went in the late 1990s, when that business model was hyped and found wanting.

By 2002, it became obvious that Loudcloud could not make a sensible living out of that business. Despite the short-term hype around ASPs, Andreessen now sees them as part of a long-term move towards outsourcing. However, it is a move that only very large companies can profit by.

“Outsourcing is a game for big companies,” he says. “Firms decide on their partner based on the balance sheet not the technology.” An outsourcing partner should be someone who will still be there in ten years, other considerations are secondary, he says.

Despite the fact that Loudcloud could not survive as an ASP, the company had built some good automation software to support its data-centres. In 2001, the company hived off its existing customers and infrastructure to outsourcing giant EDS, renamed itself Opsware and re-focussed on reselling the automation technology.

“Loudcloud had 2000 servers on three continents,” he says. “Our technology can automate and support systems from IBM, HP, Sun and others. That is how we get traction.” Outsourcers need multi-vendor utility computing schemes, instead of the single-vendor approaches of Sun and IBM. Customers include EDS, naturally, but also HP’s services arm and NEC in Japan.

“When a company outsources, the outsourcer gets 3000 servers and 3000 people,” says Andreessen. “In order to make a profit they must automate and use fewer than 3000 people, or move the work offshore.” Automation is a big enabler for the offshore stage of outsourcing, says Andreessen: “EDS is moving to Opsware as part of an offshore push. EDS is moving systems to India or Latin America but they are automated and deployed with Opsware first.”

Automation is also important to the IT manager who wants to avoid being outsourced, he says: “Ninety-seven percent of IT is still in-house.” Most companies have a hybrid approach, and efficiency is needed in-house as well.

DCML – rewriting the data centre
Opsware has got backing from EDS, and more than twenty other companies, to publish the Data Center Markup Language, an XML-based data format that defines servers and applications and other objects in a data-centre.

“Any component should be able to describe itself in DCML,” says Andreessen, a neat summation of an idea which should, if it works, remove much of the labour in current systems administration and management.

While DCML is clearly based on Opsware’s own work, and EDS’ backing is important, the other headline sponsor is a big surprise. Computer Associates, like its rivals IBM’s Tivoli and HP’s OpenView division, has a lot invested in the old labour-intensive way of building and managing data centres. “It is impressive that CA wants to change the way things work,” says Andreessen. “Tivoli would prefer that this not happen.”

Other significant supporters include the testing company, Mercury Interactive, monitoring companies NetIQ and Micromuse, the middleware companies Tibco and BEA, and utility computing company Ejasent, regarded as an Opsware rival.

Apart from EDS and CA, the big players aren’t there yet, but that doesn’t bother Andreessen, as he believes they will be forced in later. “Big companies adopt standards late,” he says, pointing out that IBM and Sun would prefer that customers buy everything from one place. “People will say ‘No way!’ to them.” Although HP is not part of the DCML group, it will be getting DCML from Opsware and BEA, he says.

Microsoft is a special case he says, since its desire to gain presence in the data centre (as well as its commitment to different models of building data centres) will overcome its reluctance to countenance a multi-vendor approach that embraces Linux. “DCML is a superset of Microsoft’s SDM [the utility-like solution Microsoft launched earlier this year]. Microsoft people would prefer the data centre to be all Windows, but failing that, this is the kind of thing they will adopt, like they adopted TCP/IP and HTML.” Ironically enough, much of those HTML developments came from Netscape, once Andreessen’s home and Microsoft’s arch-rival.

The time-table is urgent, but not hasty. A draft specification exists but will not be published until early 2004, after agreement by DCML members. “It’s being driven by [one time Netscape employee] Tim Howes, who drove the LDAP directory standard,” says Andreessen. As with classic Internet standards, the draft will be accompanied by a working, open-source reference implementation available for download. Using DCML, a server can be described exactly and cloned (which could have a direct use in disaster recovery). The description includes best practices in handling the server, and management and performance information.

“DCML will be in products in early 2004,” says Andreessen, meaning that Opsware will comply with it. He expects ten to fifteen companies to ship products during 2004. “Companies will support DCML in different ways – for instance, BEA will configure servers by feeding them a DCML description, while Micromuse could use DCML to get a description of what is there to be monitored.”

Like Netscape’s HTML work, DCML is being put to the market before the standards bodies, but Andreessen plans to offer it to a standards body (most likely the Oasis group) during 2004. Although he wants to get it into a formal standards process, Andreessen has no illusions. “The problem with the standards process is big companies have full-time standards people,” he says. “The people you want on a standard are the people who make things work.”

DCML adopts the oft-quoted slogan of the Internet standards body, the IETF: “We reject kings, presidents and voting. We believe in rough consensus and running code.”

“HTML has never been formally ratified, you know,” he says, quietly. “It’s a draft standard and keeps moving.” (Have a look at the World Wide Web Consortium’s HTML page and, surprisingly enough, the inventor of Mosaic and Netscape is right. HTML and XHTML are a series of “recommendations” based on “consensus”).

A persuasive case
Overall, Andreessen’s case is persuasive. Riding the waves of outsourcing, automation, web-enabling and cost-cutting must be a good move at this stage. One drawback could be the simplicity of the standard – sample DCML code, like other XML formats, is so “readable” it is hard to believe it is significant.

This time next year, DCML could be starting to make data centres more manageable.