Breaking the brick

Intel’s Personal Server project, lead by Ubiquitous Computing long-timer Roy Want, got some press this past week after it was shown at the Intel Developer Forum. The prototype is a 400MHz computer with Bluetooth, battery and storage, all about the size of a deck of cards. No screen and no keyboard — I/O is handled by whatever devices happen to be around, be they the display and keyboard on your desk, the large-screen projector in the conference room or your portable touch-screen. This concept isn’t new; it’s something that researchers in Ubiquitous Computing and Wearable Computing (including Roy) have been talking about for over a decade. But it is the right concept, and Moore’s Law is finally bringing it to almost within reach.

There are three main reasons why this is the Right Thing(tm):

  • Your hands aren’t getting smaller. Handheld computers are now small enough that the limiting factor is screen and button size. Since our hands aren’t getting any smaller, we’re pretty much at the limit for everything-in-a-single-brick handhelds, at least for current applications. One of the ways out of that box is the wearable computing approach, where interfaces are spread around the body like clothing or jewelry. Displays are shrunk by embedding them directly into the glasses, tiny microphones are used for speech recognition, micro cameras and accelerometers are used for for gesture and context recognition, and specialty input devices such as medical monitors get used instead of more generic input devices. One of the big difficulties with wearables is all the wires leading from the CPU/Disk/Battery unit to the I/O devices, and in fact this problem was a big motivating force behind the IEEE 802.15 short-range wireless standards, which include Bluetooth. Wireless isn’t a complete solution (you still have to worry about powering your I/O devices) but it’s a start.

    The other way to break the hand-size limit is the UbiComp approach: use whatever interfaces are in your surrounding area. When I’m at my desk I want to use my nice flat-panel display and ergonomic keyboard, not my black-and-white cellphone LCD. When I give a presentation I want to use the conference hall’s projector. I don’t need a keyboard at all, just enough to launch my Keynote presentation and change slides. Roy naturally leans towards this second approach, but as I’ve argued before the Ubicomp and Wearables approaches work well together; there’s no need to choose.

  • Always the right tool for the job. Another advantage to breaking the CPU from the I/O is it gets around an inherent conflict in interface design. On the one hand, designers will tell you that you always want the interface to fit the task. Use a hammer to drive nails and a screwdriver to turn screws, and all that. But in the mobile world you don’t want to carry around your cellphone, PDA, MP3 player, two-way pager, camera and laptop everywhere you go. When it comes to mobility, most people choose to carry a Swiss Army knife instead of a full toolchest, even though the one-size-fits-all interface won’t ever be quite right for the task. (That’s why I still carry my Danger Hiptop, which is great for text messaging but feels like I’m holding a bar of soap to my ear when I use it for voice.) When you break the brick, as it were, you can use one CPU, main battery, network connection and storage for all your devices. Then just bring whatever interfaces you need for whatever you tasks you expect that day, and use interfaces in your environment when they’re available.

  • Thin clients don’t grow with Moore’s Law. An alternative to having your personal CPU with you at all times is to run a thin client that has just enough smarts to talk to a server over wireless. The server then does all the heavy lifting. The trouble with this approach is that thin clients rely mainly on two resources: wireless bandwidth and the rather significant battery power needed to get to the nearest cell tower. The trouble is, these are the two resources that are growing most slowly. Since 1990, the RAM in mobile computers has improved a hundred-fold, CPUs 400-fold, and disk-space a whopping 1200-fold. In that same time, long-haul wireless speed has improved only 20-fold and battery efficiency only three-fold. (Thanks to Thad Starner for those numbers.) And, of course, thin clients don’t work when you’re in a wireless deadzone.

It’s not clear when Intel (or Apple or Sony for that matter) will finally come out with a successful Personal Server style product. The hardware is just one necessary piece to the puzzle, with resource discovery, communication standards, good interface design and of course the all-important “killer app” to bring it all together. But in spite of the hurdles yet to come, this is the right approach. I’m glad to see Intel is giving it the support it deserves.