Barbara Lamar wrote:
> "Scalable Routing Design Principles"
I still think that by making locally-connected networks
with nodes aware of their geographical position the best
solution is geodetic routing. Imagine the planet covered
with a slab of space containing static or mobile routers
of (time-/space-) varying density, which route via a kind
of 3d-Bresenham (actually, its higher-dimensional homologue,
since each node sees more others than the nodes on a 3d
lattice). To allow cut-through, you have to be able to
decide which space cone to choose to repeat to,
while the packet header is still streaming past you
(very short FIFO, if any), so the packet coding should
be appropriate. Of course, you should allow store-and-forward
as well, as otherwise too many packets get dropped if packet
concentration/space volume spikes too high.
For this to work, the node connectivity has roughly to
approximate the geographic proximity relationship. If
we do node-mutual time of flight "triangulation"
(scare-crowed since providing not angle but distance
constraints), the nodes can act as a global positioning
system as well as a nuke-quality time standard (if
you see a lot of high-resolution clocks, you average out
the drift). Best way to implement this would be obviously
by using forthcoming digital pulse radio, which is very
good for high-precision realtime positioning service as
well as offering excellent utilization of bandwidth, plus
compatibility with existing spectrum allocation (digital
pulse radio ultrabroadband looks a lot like white noise
background of relatively low intensity, so it doesn't jam
but degrades).
This is especially suitable to a cloud of LEO satellites
(or a large number of stratospheric platforms) with
line-of-sight laser and wireless (constant average
density of mobile nodes), as well as cellular packet
switched networks (static nodes, highest density in
urban areas). But you should be able to use TOF measurements
on legacy copper and fiber, too.
But wireless and LoS laser will get there first, obviously.
A from scratch global network doesn't have to deal with legacy
nor (overhead-riddled) packet accounting with other networks.
It could wrap legacy protocols, so both would coexist on the
same infrastructure, slowly phasing out legacy stuff as TCP/IP.
This archive was generated by hypermail 2b30 : Mon May 28 2001 - 09:56:17 MDT