The X Factor: Appliances Versus General Purpose Computers

by Timothy Prickett Morgan

Two weeks ago, I spent a weekend at a frat house at the Massachusetts Institute of Technology, one of the birthplaces of computing. Because MIT alums are serious nerds--perhaps the deadliest in the world, considering how much of the U.S. Department of Defense budget ends up in MIT labs--the frat house (pika, no initial capital letter please) held a reunion party at the Museum of Science, which is located on a bridge spanning the Charles River and separating Cambridge from Boston.

My wife, Elizabeth, went to MIT, which is how I got invited, and after getting bored with all the reminiscing by her frat brothers and sisters, I snuck away to the computing part of the museum and I actually got to examine one of the first computer appliances in the world: the Whirlwind machine, designed by MIT for the U.S. Navy as a flight simulator. No big deal, right? Flight simulators are normal. Well, the Whirlwind project began in 1945 and it was completed in 1951, and it was a big deal.

Whirlwind was the first computer to have a video display, and it was the first machine to use magnetic core memory, the predecessor to the electronic memory we have today and probably a lot closer to the nano-quantum memory our grandchildren's computers will have than today's semiconductor memory. (Mechanical computing could come around on the guitar again, once we master nano-scale manufacturing.) Whirlwind was also noteworthy in that is was the basis of the Semi-Automatic Ground Environment (SAGE) computer system that was awarded to IBM and which made it a player in the electronic computing game; IBM was still a punch card company at this time. Segments of the SAGE system were operational until 1983, by the way, and that has to be a record of sorts for the computer industry. And it is probably only the lack of vacuum tubes from the United States--Phillips shut down its last plant in Altoona, Pennsylvania several years later--that forced Uncle Sam to finally pull the plug on SAGE.

Whirlwind was also important for another reason: one of the engineers on the project was a guy name Ken Olsen, who would watch the mainframe market develop as he tooled around at MIT in the Lincoln Lab and then, with a bunch of Brass Rats, would initiate the minicomputer revolution with his founding of Digital Equipment Corp and the delivery of the PDP-1.

(I have decided to call MIT grads this nickname; they do not refer to themselves as this. I am referring to the gold MIT ring graduates get, which has a beaver on it--Nature's Engineer--and which MIT grads call a Brass Rat. I think it is a good nickname for MIT grads. And fifteen years from now, after I have got IT Jungle and other things humming along, I am seriously thinking about applying to MIT and finishing my engineering degree. My wife has the same dream. I want a Brass Rat of my own, and she wants to finish getting hers, too, after deciding to jump to the University of Chicago and getting bachelor's degree in medieval history. I studied aerospace engineering for five semesters at Penn State before jumping to an English major, and I think I made the right move, but not having finished that engineering degree still bothers me. We both like engineers, and we still want to be engineers. I want my children to be engineers. Not because it is a great job in terms of economics, but because it is important to be a problem solver and a builder.)

One of the amazing things about Olsen is that he was working for IBM at the time he and his compatriots at MIT were conjuring up the idea of a minicomputer with standardized components. He made a proposal for creating this new market to Big Blue, and IBM said it didn't see any merit in the idea. That's about as dumb as telling Ross Perot, the top mainframe salesperson in IBM's Wall Street area, that the idea of a services business was not worthy of its attention; he went on to found Electronic Data Systems in 1962, and it is now a $20 billion company and the main reason why half of IBM's sales are now coming out of its Global Services unit. Any idea that IBM summarily discards is probably a good one.

The thing that struck me about Whirlwind as I was standing there in the Museum of Science, aside from its size and vintage, was the fact that it was not really a general purpose computer at all, but rather an appliance. That may sound odd as a description for a computer that took up 3,100 square of space and that had 4,500 vacuum tubes, but when you get right down to it, Whirlwind was an appliance that was supposed to be a flight simulator but which ultimately was turned into an appliance to watch the skies for incoming bombers and missiles.

As I stood there, I couldn't help but wonder: Why aren't all computers these days appliances? Like an Apple iPod. Or a Sony PlayStation. Or a Nintendo GameBoy. Or a toaster. But for heaven's sake, not like the VCR that no one knows how to program after two decades of supposedly improved interfaces.

This may seem like a stupid question, but think about it for a second. When we buy general purpose computers--like Unix or Wintel or Lintel servers in our data centers or the PCs on our desktops--what do we do? We install our specific applications on them and we turn them into our own special appliances. The hardware and software are general purpose in that the resulting system can, in theory, run many different kinds of applications and have many different settings in terms of connectivity and user preferences, but ultimately, we create an appliance. However, we have to suffer through all of the complexity associated with a modern general-purpose computer--on that has an operating system that can, in theory, run almost anything. This, it seems to me, is not a fair trade.

Real appliances have gained a little traction in recent years. For a while, before Sun Microsystems bought it and killed it, Cobalt Systems was proving that a low-cost, high-volume, Internet-server-for-dummies appliance was appealing to the market as an alternative to rack-mounted X86 servers. Network Appliance proved that Ethernet-attached, network-accessible, generic disk storage was something customers would want over the complexity of installing shared storage area networks for their server farms or relying solely on direct-attached disks for their servers. And there are a slew of infrastructure appliances, which do firewall, Web caching, search engine, print serving, and other functions. The neatest new appliance to come along in a while is the Java Compute Appliance from Azul Systems, which is comprised of server blades with a 24-core processor in each blade that are designed, from the transistors up, specifically and only to run Java Virtual Machines. Azul is offering excellent Java performance for a fraction of the cost of running Java on generic servers.

To be an appliance means two things: ease of use and precisely tailored function. It can mean lower cost than general purpose computers as well--as is the case with infrastructure caching servers or the new compute appliances from Azul. But this price differential is not a requirement. A standard iPod is more expensive than the best Walkman ever was--even at the peak of the Walkman's popularity--and the iPod performs essentially the same function--albeit with better technology and a better interface. But what makes an iPod really useful is that it is an appliance that is a lot less difficult to use than a PDA or cell phone or a laptop. It is a tool that does its job, not a tool that promises to be any tool you need. And in many respects, this is a better design method than the general purpose machines we have in the data centers, or have on our desks or lug around on our backs every day.

What we might need are real seal-box appliances. Here is an Oracle database appliance with the SAP database hard coded, and here are 12 SAP application appliances and a few infrastructure servers. They are set up at the factory, and you don't get to mess with them. Let's face it: You don't want to be an SAP performance expert anyway, and if these things are mass produced, you just won't care. You don't want to build a data center from scratch using generic machines so much as install a collection of appliances that not only run your applications, but for all intents and purposes are your applications. If you need to add some more users, you add another application appliance, which might be rated for 50, 100, or 500 users. If you need more database performance, you buy a bigger appliance or you add one to the database appliance cluster and it sorts out how to share the work because it is an appliance and that is what appliances do.

And the same goes for the desktop. Instead of having a big, wonking 3 GHz, 64-bit chip and a generic operating system, what if your desktop was a high-speed parallel supercomputer backplane and a clustering layer of APIs that allowed multiple, very small and low-powered appliances (using something like 1 GHz Eden chips from VIA Technologies) to rapidly communicate and simulate the multitasking environment? You would buy an office application suite appliance (perhaps with Microsoft Office, perhaps with StarOffice, on it), an Internet appliance (with a browser and network connectivity), a print and file server appliance (for linking to printers and storing documents), and maybe a graphics appliance (with varying levels of video performance). Each appliance would be tuned to give a certain level of performance. If you wanted to spend more money, you could get a more powerful appliance.

As one element became obsolete in this appliance world, you could trade them out. Instead of having to upgrade your servers and messing around with software licenses and all that, you pull out a bunch of appliances and you put in a bunch of new ones.

For all the rigidity that would seem to be apparent in opting for appliances over general purpose computers, this clustered appliance approach could end up being a lot less costly and a lot more flexible.