Thesis 60


Thesis 60

The necessary storage capacity already exists.

It's easy to infer that a panoply of ubiquitous systems running at all timessystems whose operation by definition precedes users, as we've notedis going to churn up enormous quantities of data. How and where is all this information going to be stored? Will the issue of storage itself present any obstacle to the real-world deployment of everyware?

We can derive a useful answer by, again, extrapolating not from the best currently available systems, but from those at the middle of the pack. The iPod shuffle I wear when I go running, for example, is a circa-2004 solid-state storage device, with only incidental moving parts, that boasts a capacity of 1 GB. This is about a day and a half's worth of music encoded with middling fidelity, a few hours' worth at the highest available resolution. It achieves this (as Apple's advertising was pleased to remind us) inside a form factor of around the same volume as a pack of chewing gum, and it's already been rendered obsolete by newer and more capacious models.

A day and a half sure sounds like a decent amount of music to pack into a few cubic centimeters; certainly it's suggestive of what might be achieved if significant parts of a structure were given over to solid-state storage. But hard-ubicomp enthusiasts already dream of far greater things. On a chilly night in Göteborg in late 2002, Lancaster University HCI pioneer Alan Dix described an audacious plan to record in high fidelity every sense impression a human being ever hasfavoring me with a very entertaining estimate of the bandwidth of the human sensorium, the total capacity necessary to store all of the experiences of an average lifetime, and a guess as to what volume would suffice to do so: "If we start recording a baby's experiences now, by the time she's 70 all of it will fit into something the size of a grain of sand."

If I recall correctly, Dix's order-of-magnitude guess was that no more than 20 TB (each terabyte is 1,000 GB) would be required to record every sensory impression of any sort that you have in the entire course of your life. And when you run the numbersmaking the critical assumption that increases in storage capacity will continue to slightly outpace the 24-month doubling period specified by Moore's law for transistor densitymirabile dictu, it does turn out to be the case that by mid-2033, it will at least theoretically be possible to store that amount of information in a nonvolatile format the size and weight of a current-generation iPod nano. (The grain of sand appears not long thereafter.)

As of the end of 2005, the numbers undernetting this rather science-fictiony-sounding estimate still hold and are maybe even a little conservative. The real point of all this extrapolation, though, is to buy some room for those challenges inherent in everyware that, however daunting they may seem at the moment, are nonetheless of smaller magnitude.

If we can take as a limit case the recording of every single impression experienced in the course of a life, then it seems fair to say that all the other issues we're interested in addressing will be found somewhere inside this envelope. And if this is soand there's currently little reason to believe otherwisewe can safely assume that even devices with small form factors will be able to contain usefully large storage arrays.

Going a step further still, such high local information densities begin to suggest the Aleph of Borges (and William Gibson): a single, solid-state unit that contains high-fidelity representations of literally everything, "the only place on earth where all places are." As strange as this poetic notion may sound in the context of an engineering discussion, the numbers back it up; it's hard to avoid the conclusion that we are entering a regime in which arbitrarily large bodies of information can be efficiently cached locally, ready to hand for whatever application requires them.

If this is too rich for your blood, Roy Want, Gaetano Boriello, and their co-authors point out, in their 2002 paper "Disappearing Hardware," that we can at least "begin to use storage in extravagant ways, by prefetching, caching and archiving data that might be useful later, lessening the need for continuous network connectivity."

While this is more conservative, and certainly less romantic, than Borges' Aleph, it has the distinct advantage (for our immediate purposes, anyway) of referring to something real. Intel has demonstrated several iterations of a high-density mobile/wearable storage system based on these ideas, called a "personal server," the earliest versions of which were little more than a hard drive with a built-in wireless connection. Where Want's version of Alan Dix's calculation puts that total lifetime throughput figure at a starkly higher 97 TB ("80 years, 16 hours a day, at 512 Kbps"), he reckons that a personal server should store that amount of data by the more optimistic date of 2017; some of the apparent optimism no doubt reflects the difference in scale between a grain of sand and the mobile-phone-sized personal server.

But, again, the purpose of providing such calculations is merely to backstop ourselves. Any ubiquitous application that requires less in the way of local storage than that required by recording every sensation of an entire life in high fidelity would seem to present little problem from here on out.