November 18, 2002 (InfoWorld) --
Steering his company from a managed services business model into uncharted software provider waters, Marc Andreessen, chairman and co-founder of Sunnyvale, Calif.-based Opsware Inc., is no stranger to unique challenges. The former chief technology officer at America Online Inc. and Netscape Communications Corp. recently discussed his thoughts on IT automation, the rapid commoditization of hardware and why blade servers are ultimately headed for the trash bin.
What are the enterprise and budget consequences of current IT automation solutions focused on cutting costs? The consequences are really good. There're a whole bunch of interesting changes going on as a result of lowering cost. People are going after hardware with a meat cleaver -- they're trying to figure out how to radically reduce costs of hardware. We're seeing that across the board. We're seeing a very broad pickup in Intel-based servers for enterprise applications vs. more traditional proprietary big iron, including big Unix systems.
The way people look at that is they're going to switch from having a $300,000 big Unix box to having 10 $3,000 Intel servers, and they'll be running the exact same application, and they'll be running it faster, and it'll be more redundant and a lot cheaper.
What are Opsware's plans for the blade server space? In general, blade servers are a subset of a broader trend we see happening that we call "disposable servers." Servers have gotten a lot cheaper since the mainframe and since the '80s. A server two years ago was likely to be a box costing anywhere between $30,000 and $3 million, and it was a big, important box that runs big, important applications, and if it breaks or something goes wrong, there are a lot of people that are going to get involved in fixing it. If people shift to the distributed application architecture with the Web, application servers and redundancy, and [with] horizontal scaling where they'll tend to run a much larger number of servers but those servers will tend to be much cheaper, you head into a world where the servers themselves become disposable, and I literally mean disposable. Which is, a server that costs $1,000 or $2,000 is not worth fixing. It's so cheap, it's not worth the effort to ever try to open it up and fix it. If it breaks, if the hard drive crashes or CPU crashes, you throw the server away.
That can apply to Intel [traditional] rack-mounted servers, or the same concept applies to blades. If a blade failed, you're going to throw the blade away and pop a new one in. And that's a big, big change, and it's going to really drive a lot of cost reduction. ...
The consequence of that is you're going to have a much larger number of servers that are individually going to be much cheaper and disposable, and applications will be written to be redundant across many servers so you can fail-over and can throw those things away.
How can customers align their hardware and software investments with business needs? Right now, when you think about IT, every application has its own infrastructure and its own servers -- everything is very ad hoc, and people are running around fighting a lot of fires. In the future, I think you'll think about it more as capacity on demand: File a formal request for new capacity or system -- those come online very, very quickly -- [and] there's a formal service-level-agreement relationship between the IT department and the business user, and that's tracked and managed by software. There's a formal billing relationship, where you know how much this stuff is costing. There's the ability to scale up and scale down very smoothly to make sure that the cost matches what's actually needed.
Reprinted with permission from For more enterprise computing news, visit Infoworld.com Story copyright 2006 InfoWorld Media Group, Inc. All rights reserved.
See results from our survey of more than 5,000 IT pros, and use our Smart Salary Tool to compare your pay with IT workers in similar jobs across the U.S.