::scr Seti@Home & Global Warming

David Cantrell scr@thegestalt.org
Mon, 20 May 2002 12:27:47 +0100


On Mon, May 20, 2002 at 11:29:33AM +0100, Simon Wistow wrote:
> On Fri, May 17, 2002 at 02:05:20PM +0100, David Cantrell said:
> > Not true.  For *some* types of calculations, it is the most efficient way
> > of doing them.
> Ah, slip of the forked tongue (he says pausing briefly to revive the
> horse using CPR before swiftly flogging it to death again) - I meant
> it's not efficent energy wise since there's he extra power needed to
> pump it over the wires, run the switches and routers and stuff etc etc.

Ah, right.  In that case, yes, you do have a point.  However, I still
maintain that there is a set of problems which could very well be more
efficiently done overall in this way.  Again, taking the example of OGRs,
you need to do a VAST number of relatively simple operations on each
candidate to figure out if the GR really is O.  That parallelises really
well.  It also requires virtually no network traffic (basically, the
server sends a smidgeon of information telling the node what to work on,
the node sends back a smidgeon of information hours or days later saying
which candidates are either definitely not O, definitely O, or neither
(so need more work).

The amount of power used to transmit those few packets is minute compared
to the power taken by the processor on the node.  Yes, there is a small
amount of extra energy used.  But compare that to the purchase price and
running costs involved in building and maintaining a single machine which
can do the same number of Hard Sums in the same amount of time.

Unfortunately, by the time your centralised facility has saved enough
energy to have a measurable effect, the decentralised network will have
grown further, have faster nodes doing more sums cheaper, and so still
come out ahead.  Until, of course, the network grows so large as to
over-stretch the capacity of the servers.  Even that can be put off for
a considerable time by using a distributed network of servers.

> Then I was thinking about the energy required to do calculations - I
> once had cryptography explained to me in which the person showed that
> for a certain key length (not even that long IIRC) then if we built a
> dyson sphere round a star (http://www.d.kth.se/~asa/dysonFAQ.html) and
> lined the inside with switches then there wouldn't be enough power in
> the star to flick the switches enough times to do every permutation of
> binary bits in order to crack the key.

That assumes that the only way to crack it is brute-force.

> Conversely, I think I'd be right in saying that the more energy that you
> chuck at a problem, the faster you can solve it.

I don't think so, but I'm having trouble putting it in to words.  But
think about:
  heat dissipation (remember, energy is never destroyed, it just gets
    turned into heat)
  communications latency across large pieces of equipment because of
    those pesky Michaelson, Morley* and Einstein chaps.

At some point though, using more and more energy will not only give you
diminishing returns, but also will probably slow you down.  If you want
to solve problems faster, the best way will always be to build a better
algorithm.

> we could solve all problems pretty much instantaneously and then sit
> around drinking daquiries wondering why the computer spat out the answer
> 42.

A quantum copmuter would do that without needing large amounts of energy.

* - M and M are the pair who put an accurate measurement on the finite
speed of light

-- 
Lord Protector David Cantrell     |     http://www.cantrell.org.uk/david

   Lefties are usually well-intentioned at least, and possess
   a far greater command of grammar and spelling.
      -- Noel, in soc.history.what-if