RANT: wiring up the neighbourhood

Eugene Leitl (Eugene.Leitl@lrz.uni-muenchen.de)
Sun, 13 Jul 1997 19:37:42 +0200 (MET DST)

Kauffman nets (randomly wired graphs) show a very interesting
property: if the ratio of edges/nodes goes over the threshold
of 0.5, most nodes crystallize into one giant, interconnected
domain. A classical phase transition, obviously: gradual
change, before, gradual change, after; a very rapid change,

This has some impacts onto the grassroot nets I've been
proposing for years (seasoned denizens will probably remember.
Several people came up with related schemes, it is probably a
FAQ item somewhere). But let's describe the status quo first, to
learn what exactly what, if anything, is amiss with the way we
are accustomed to do things just now.

Currently, there is a sharp boundary between the consumer and
the provider. Single users have to pay the telcos for use of
their comm lines, the providers for bandwidth and computational
resource. In some, less advanced regions of the world, a telco
monopoly still prevails. Behind it, hides the ISP semimonopoly,
attempting to maximize profits by keeping newcomers out. It
won't hold for long for sure, but damage's done, already.

On the other side of the spectrum, bottom-up networking has
begun. We have the CAN bus on the side of the microcomputer
embeddeds, IBM's PAN via the "handshake" protocol, ethernet
networking in the flat; between home PCs, IrDA (now extended to
4 MBit/s) to talk to the portable, wireless modems, and the
like. With the recent advent of cheap multiport cards, 10/100
switches, GBit ethernet, and plastic fiber optics relatively
impressive hardware will become available to the home user.

In a very real sense, random interlinking has already begun. It
is possibly only a question of time, before adjacent domains
crosslink and coalesce, resulting in a larger nucleus. Its
participants profit mutually, as a linkup has emergent
properties (distributed storage, resource sharing, etc.) If we
extrapolate this way into the future, we'll get a global
grassroot net.

A quick sketch, how this can be possibly accomplished.

Birds of a feather flock together. This also applies to likewise
minded individuals. In some regions the critical potential user
density is already reached. A (possibly boosted) IrDA module, a
pair of magnifying bathroom mirrors (to collimate the 30 deg
beam), and a PPP infrared link across the street is in place.
Links along the street may form a pretty good backbone, ranging
for many 100 m even with primitive optics. There are few
problems assotiated with a twisted pair network in the house
(current loops), put galvanic potential-separating cheap optical
fiber is around the corner. Using laser diodes and directed
microwave links will be the next step.

Many users object to let their machines run 24 h/day, also most
OSses show too limited performance and availability (two
free Unix users won't probably find any of this very demanding).
A hardware solution is in order, which may well be based on a
dedicated embedded solution, or on embedded PCs running Linux,
or, possibly, (embedded) Java in the near future. A low-cost
solution would use (optocoupled, or optical (IrDA)) RS232, a
more expensive one, ethernet. The system must be performant
enough to be able to handle several channels simultaneously (in
some cases a simple crossbar may already suffice, in some
interim storage and packet checking might be mandatory). The
system must come in a rugged, atmosphere-proof case, large
enough to contain a buffering battery (in case the system is
powered by a solar cell, crucial for a standalone relais).
The basic idea here is that users own the hardware, and do some
very limited maintance.

There are several things to tackle: nanocash packet accounting
(to prevent network abuse), firewalling (agents should be able
to live in the boxes, but not to enter your home machine & to
do some havoc there), and a geodetic routing design.

The latter point is tough. Ubiquitous networking will need a
very different routing scheme than current TCP/IP. Constraints
like visibility demand such a scheme (tesselating space in
addressable voxels, which can be occupied by a hardware box).
The latter demands a huge address space. If you can imagine a
mm^3 voxel sphere ranging to beyond GEO, you might notice that
IPng has got a problem (the raw address space would suffice).
In the routing scheme there should be no difference between
routing and switching. Boxes, knowing their spatial address
(from GPS, and/ultrasonic, or vis LPS), can select the nearest
one. The benefits are twofold: you can't fake your address,
since your neighbouring boxes would notice the tampering, and
won't route your packet. Additionally, this gives a great means
for automated traffic routing, whether for humans, or agents.
(Remember a recent aerobot thread).

This could go on for hours, I obviously have to truncate. Such a
routing scheme appears devisable, using orthogonized hypercubes:
hypergrids, or, lattices. Their capability to handle wraparound
at address space boundaries fits nicely with curved spheres,
both of surface, and orbital configurations. With some effort, a
fuzzy routing scheme probably could be devised, which derives
fuzzy addresses from wiring constraints.



P.S. I've got a related problem: I'd like to wire up with
friends of mine living in the vicinity to share an ISP backlink
to be able to use cable modems, which are unidirectional (all
thanks to my favourite monopoly (Telekomiker)).

P.P.S Just hooked up a tiny PC keyboard to my Newt130. The Newt
(especially the Newt2000) is imo a pretty interesting minimal
wearable platform, since both offering computational oomph (the
NTK, a sofware developer set is available for free for download,
and comes bundled with many Newton developer books, it allows
the use of NewtonScript, reminiscent of Java, and native code
as well), and is not too demanding in terms of power burned).
Of course the peculiar demands of a wearable (which are weird,
and many) demand a hackbox.