On Oct 5, 2013, at 2:00 PM, John Gilmore wrote:
b. There are low-end environments where performance really does
matter. Those often have rather different properties than other
environments--for example, RAM or ROM (for program code and S-boxes)
may be at a premium.
Such environments are getting very rare these days. For example, an
electrical engineer friend of mine was recently working on designing a
cheap aimable mirror, to be deployed by the thousands to aim sunlight
at a collector. He discovered that connectors and wires are more
expensive than processor chips these days!...
He had a big advantage: He had access to power, since the system has to run
the motors (which probably require an order of magnitude or more power than all
his electronics). These days, the limits on many devices are expressible in
watts/compute for some measure of compute. But less often noticed (because
most designers are handed a fixed device and have to run with it) dynamic RAM
also draws power, even if you aren't doing any computation. (Apple, at least,
appears to have become very aware of this early on and designs their phones to
make do with what most people would consider to be very small amounts of RAM -
though perhaps not those of us who grew up in 16-bit-processor days. :-) Some
other makers didn't include this consideration in their designs, built with
larger RAM's - sometimes even advertising that as a plus - and paid for it in
reduced battery life.)
A couple of years back, I listened to a talk about where the next generation of
wireless communications will be. Historically, every so many years, we move up
a decade (as in factor of 10) in the frequencies we use to build our
communications devices. We're approaching the final such move, to the edge of
the TeraHertz range. (Another factor of 10 gets you to stuff that's more like
infrared than radio - useful, but in very different ways.) What's of course
great about moving to higher frequencies is that you get much more bandwidth -
there's 10 times as much bandwidth from 10GHz to 100GHz as there was from DC up
to 10GHz. And the power required to transmit at a given bit rate goes down
with the bandwidth, and further since near-THz radiation is highly directional
you're not spewing it out over a sphere - it goes pretty much only where it's
needed. So devices operating in the near-THz range will require really tiny
amounts of power. Also, they will be very small, as the
wavelengths are comparable to the size of a chip. In fact, the talk showed
pictures of classic antenna geometries - dipoles, Yagi's - etched directly onto
chips.
Near-THz frequencies are highly directional, so you need active tracking - but
the computes to do that can go on chip along with the antennas they control.
You'd guess (at least I did until I learned better) that such signals don't
travel far, but in fact you have choices there: There are bands in which air
absorption is high, which is ideal for, say, a WiFi replacement (which would
have some degree of inherent security as the signal would die off very
rapidly). There are other bands that have quite low air absorption. None of
these frequencies are likely to propagate far through many common building
materials, however. So we're looking at designs with tiny, extremely low
powered, active repeaters all over the place. (I visualize a little device you
stick on a window that uses solar power to communicate with a box on a pole
outside, and then internally to similar scattered devices to fill your house
with an extremely high speed Internet connection.)
The talk I heard was from a university group doing engineering
characterization - i.e., this stuff was out of the lab and at the point where
you could construct samples easily; the job now was to come up with all the
design rules and tradeoff tables and simulation techniques that you need before
you can build commercial products. They thought this might be 5G telephone
technology. Expect to see the first glimmers in, say, 5 years.
Anyway, this is (a) a confirmation of your point that computational elements
are now so cheap that components like wires are worth replacing; but (b) unlike
the case with the mirror controllers, we'll want to build these things in large
numbers and scatter them all over the place, so they will have to make do with
very small amounts of power. (For the first inkling of what this is like,
think of RFID chips - already out there in the billions.)
So, no, I don't think you can assume that efficiency considerations will go
away. If you want pervasive security in your pervasive compute architecture,
you're going to have to figure out how make it work when many of the nodes in
your architecture are tiny and can't afford to drain power run complicated
algorithms.
-- Jerry
___
The cryptography