[extropy-chat] sjbrain calcs

[Eugen Leitl]

On Mon, Dec 22, 2003 at 01:06:44AM -0700, Alan Eliasen wrote:
 
>    Can I ask what may be a silly question?  Why would one build so many
> little far-flung nodes instead of a more monolithic structure?  So many

This isn't a silly question at all. I agree with you that the designs Spike
has been floating are way too dilute and too gossamer. This starts even with
orbit control: you need to be able to rearrange mass in order to tilt the
sail (reaction mass is right out). You can't rearrange anything in a flat 1
cm disk 1 um thin (allright, you can tile it with electrostatics-actuated
tiltable MEMS mirros).

> things are simpler, faster, and more efficient with a single
> structure--communications, energy expenditure, time lag, that I guess I
> don't see why the structure wouldn't be made much more compact.

There are several limits at play here. You correctly state that one would tend
to have a compact assembly of switches to minimize lag. However, there's
another limit: you have to be able to cool it (even given largely reversible
computing and fractal cooling channels, that's probably something in the
cubic meter range at best). Hot circuitry runs at up to ~600 K, but some
type of computing circuitry likes it cold -- very cold, few K. Of course you
can recapture the leaks and the IR blackbody of the inner device clouds by
a yet another cloud farther out. Such an object would be impossible to
detect, since radiating some ~MT of energy/s but at few K and many light-min to
light-hours across. 
 
>    For example, let's be way overgenerous and assume each tiny node had
> the computational power of a human brain (probably not possible with
> only 50 billion atoms) and that it switched at a rather conservative 1

Let's say a switch+overhead takes 10^3 atoms. This is not conservative at
all, and in fact rather tight. 50*10^6 switches is an insect, not a human.
To pull a number from /dev/ass, I wouldn't go below a cm^3 for a node (and
not above a m^3, unless it's running in a mode very close to fully
reversible, so power dissipation is negligible). 

It takes a sail/panel several m^2 to feed it, and to provide enough photon
momentum to maneuver, and you'll need some 100 m elbow room between the
nodes.

> THz.  (We'll assume a classical architecture; it gets worse for
> smarter/faster quantum computers.)  This is approximately a factor of a

Don't assume QC, but it's safe to assume reversible computing at least in the
outer, cold device cloud.

> billion times faster than the fastest neurons in the human brain can
> switch.  We'll oversimplify and say that each node thus "thinks" a
> billion times faster than a human brain.

I wouldn't postulate 10^9, but 10^6 speedup is conservative, imo. Basically
you're having a ~ns time scale for processes which take ~ms in biology. ~ps
switching at high device density and nontrivial operation complexity is 
probably pushing the envelope a bit.
 
>    Now, I've talked to people on the other side of the world on the
> phone, and the satellite lags were very noticeable and confusing.  We'd

Relativistic lag takes getting used to. I've had lag up to a minute on VoIP,
now that was really confusing (especially, if the lag varies, and you don't
know how long it is right now)

> pause too long, interpret silences as reticence, or feel we had said
> something wrong, and then both people would start talking at the same
> time, only to interrupt each other a half-second later, and then stop
> and start over again.  And that's a fraction of a second delay.
> 
>    Communicating with someone with a one-minute lag would probably be
> pretty inefficient and painful.

It is very doable. In practice, there will be clustering of units which need
to talk frequently, the algorithm to compute next direction of the hop is
very simple.
 
>    Now, when you have a node that's a billion times faster, and you're

Let's say a million.

> communicating over even a light-second distance, that's an infinity.  It

A ns is 0.3 m, a us is 300 km. There's plenty of resources within that
volume.

> might be like a billion seconds to a human.  A billion seconds is over

At a million speedup a day is about 3 kYrs.

> 31 years.  It might seem that long to a node to ask a question and get
> an answer.  It's like carrying on a conversation with
> 
> Using Frink notation, ( http://futureboy.homeip.net/frinkdocs/ ) *
> 
>    1 billion seconds -> years
>    31.68876
> 
>    When communicating over a light-minute radius, a node would grow old
> and bored and completely forget the question.  Or at least I would.
> It'd have the responsiveness of an IP-via-carrier-pigeon Internet, which
> was actually proposed:
> 
> http://www.faqs.org/rfcs/rfc1149.html
> 
> Data:   "And for a time, I was tempted by her offer."
> Picard: "How long a time?"
> Data:   "0.68 seconds sir. For an android, that is nearly an eternity."
> 
>    So, my question is, why intentionally make it so hard and painful to
> communicate?  I've always thought that future supercomputers would
> cluster as close as possible together so that communications lags would
> be minimized.  Or, why not?  Robustness?

Current dies dissipate more than a heating plate, and are within touching
distance of nuclear reactor core power density. A circumsolar cloud of
computronium marbles in low orbit can be modeled as a blackbody with a low error margin
(because each part of the cloud is basically in equilibrium). 

It doesn't include the sails for active orbit control, but what the heck.

P.S. I had a number of posts in postponement, but I just don't have the time.
Apologies for those who expect an answer for me; they might or might not yet
be forthcoming. I will unsubscribe from a number of nonessential lists, so I
won't be very visible in the new year.

Nice holidays to you all.
 
-- Eugen* Leitl <a href="http://leitl.org">leitl</a>
______________________________________________________________
ICBM: 48.07078, 11.61144            http://www.leitl.org
8B29F6BE: 099D 78BA 2FD3 B014 B08A  7779 75B0 2443 8B29 F6BE
http://moleculardevices.org         http://nanomachines.net
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 198 bytes
Desc: not available
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20031229/13d16c36/attachment.bin>