Consider the telephone answering machine. It began as a bulky analogue box running spools of tape. It turned into a small digital box, often incorporated into a phone. And finally it disappeared altogether, turning into pure software running out somewhere on a phone company’s network. Once you bought an answering machine. Now you buy an answering service.
And so it goes. Software kills hardware. We give it a fancy name – “virtualization” – but it’s just a matter of programming a computer to do what used to require a separate appliance. And as the computer gets more powerful, more and more appliances get sucked into its software.
In a post on Monday, Sun’s Jonathan Schwartz noted how the people who run corporate data centers have had a love affair with special-purpose appliances: “NAS filers, load balancers, storage switches and firewalls, even custom search appliances.” The appliances “solve a specific problem, do so with great focus, and are like novacaine on a technical problem. Have a pain? Numb it with an appliance.” But there’s a high price. The proliferation of specialized gear quickly becomes an economic and operational burden: “Leaving high price tags aside, specialized products typically require specialized skills, customized management or versioning processes, and they tend to be difficult to integrate into increasingly uniform datacenter processes.”
Fortunately, all that stuff is going the way of the answering machine. The functions are being programmed into computers, which not only saves money but also increases flexibility. Software’s a hell of a lot more malleable than hardware. Schwartz quotes one forward-thinking customer who says that “general purpose [computers] are so fast, we do pretty much everything in software.”
Where does it end? I was talking recently with Bryan Doerr, the chief technology officer of the hosting company SAVVIS and a guy who’s thought a lot about the implications of virtualization. He suggested that eventually we’re likely see the arrival of what he calls “the virtual data center.” You’ll be able, in essence, to encapsulate in software the configuration of an entire corporate data center. Need to set up a new center? Just run the program.
And, of course, once the data center turns into software, you can automate its operation and management. And you can set it up wherever you want – on your own computer or on somebody else’s. In the end, it probably just gets sucked into the network. Like the answering machine.
Software kills hardware. Think about it.
Software kills IT managers
Software kills hardware. Think about it.
I’ve thought about it, and I think all that software is going to need some hardware to run on. What you’re really saying is that software kills specialised hardware – or perhaps, more interestingly, that software kills hardware units, till the day when everything is running on one big lump of virtualised hosting power. I’ve always thought the Web was tending towards the condition of Asimov’s Multivac – only with the intelligence of Wikipedia and various folksonomies…
What you’re really saying is that software kills specialised hardware
Correct.
Software kills IT managers
Let’s just say it frees them to pursue other opportunities.
I really hope that was a Vonnegut reference.
Nick,
Bryan Doerr is right. The target is a fully virtualized utility that runs web applications on a flat grid of commodity servers. No firewalls, load balancers or NAS boxes. No complex and expensive SAN. Just rack after rack of commodity servers connected by commodity Ethernet.
This utility runs multiple N-tier applications on the same grid. It can scale each application up or down dynamically with demand, from a fraction of a server to hundered of servers. It makes it easy to add and remove hardware on the fly. It tolerates hardware failures gracefully and without losing data. And it is easy to operate and manage, because it takes care of itself in most ways, leaving the operator to manage the important aspects.
At 3Tera, we have built the software system that makes this utility work. The product we have today already does everything I’ve listed above. What’s more, it allows you to take your existing stateful N-tier web applications and move them to the grid without any code changes.
It is already in use by more than a dozen SaaS and Web 2.0 companies, and we are working with infrastructure providers (Savvis included) to make it available to the general market. For a sneak preview, see our online demo at http://3tera.com/demo/online-demo.html
Nick
12 years ago I had a meeting with one of the Human Resource executives at Nokia. We were talking about developing the right skills set for competiveness in the 21st century.
She told me that Nokia’s biggest challenge was to become a software company too. In the early 1990’s they had hired people that could build hardware, but they soon realised that software was the key. She was right.
The company who seems to live this idea to the fullest is Google, where all internal hardware seems to be cheap and generic.
All the more amazing that their Enterprise product (the Google Search Appliance) is the complete opposite of this – proprietary hardware for a single purpose – powering a search engine for company-internal data/information, which in itself sounds more like a software task.
What’s the reason for this? I bet it’s the ease of deployment and support you can accomplish through this model – if you control the entire operating environment, you minimize variabilities and thereby setup times and support incidents.
But do you really need proprietary hardware for this? Maybe technology companies should deploy their products as software embedded in “virtual appliances” (a VMWARE term), completely preinstalled in a single-purpose LINUX-based operating environment. This should give you the best of both worlds…
That reminds me of a quote I’ve heard attributed to one of the team that designed RAID disk drives. “Why did we build RAID? Because all hardware eventually breaks. All software, on the other hand, eventually works.”