incidentally
| > In particular I've been wrapping my brain around thinking
| > of Vadim Antonov's approach of a "stacked network" of
| > several identical interconnected planes, where each plane
| > is a set of vertices (routers) interconnected by lines
this doesn't say it explicitly, but this appears to be a solved
problem. it has serious practical scaling problems in terms of
state discovery, which i'm sure you're well aware of. :-)
(i.e., the best you can do is to grow ls exponentially within
a plane and geometrically plane-to-plane, if you eliminate
viable (but long) paths from consideration at the outset.
that is, you get into the meat of the vertex abstraction
problem without touching the line abstraction problem.
also, uuwho's network is bowl-of-spaghetti, whereas vadim's
proposal was the familiar square-with-cross neatly stacked
multiple times. oc3 pos first then gsr next meant we never
rolled this expensive thing into the real world. lothberg's
toy is neat too, as a 3-fold stacking of the entire current
sprintlink architecture ("super pop"), to forestall the worsening of
their current spaghetti-problem, and also as a 3x3x3 srp
"hypercube" ("hyper pop") giving 48 x 2.5Gbps access rings
and sufficient capacity in/out of the collection of routers, wheeeee
personally, i'd like to consider avoiding the spaghetti-vs-overbuild
tradeoffs in the absence of point-like forwarding gear and faster
single lines, hence my initial question...
Sean.
This archive was generated by hypermail 2b29 : Mon Aug 04 2003 - 04:10:04 EDT