|
North American Network Operators Group Date Prev | Date Next | Date Index | Thread Index | Author Index | Historical Re: Optical Crossconnects and IP
[...] > > I both fly (although tiny stuff only and otherwise only simulations ;-) > and do the networks stuff so here's a little comparison: > > Your airplane metapher almost works BUT if you talk about running even 300-400 > nodes with a link-state, you should compare it to flying an F18 formation of 30-40 > airplanes. A single one can turn well, yes, but try to get a coordinated, standard rate, announced > speed turn of 30 aircrafts and you start to understand the dynamics. There is a theory > that deals with that stuff (it's called in german "Schwarmtheorie", english equivalent to "swarm" > or "flock" probably) and it speaks about thing like center of a flock and it's mass ASIR but I never > saw it applied to networking, funny enough. So my point being, it's not like single-mass > Newtonian physics, > you basically get into the problem of Laplace grids (I think that's what the grids of masses > with springs inbetween were called ;-) and changing traffic characteristics are probably equivalent > to kinetic energy pumped into the system from its edges. > That's where your differential equation systems start to look > hard and pretty soon you're on the border of chaos theory. Somebody in Santa Fe should start > looking into that ;-) > [...] > > -- tony > Your point being that I have grossly underestimated the difficulty. A 20x20 grid transformed out of LaPlace space should generate thousands of significant partial differentail terms, then add the chaotic and discontinuous nature of the input. Gee, I wonder why the feeback loops are hard to stablize in networks. Add yet more by the fact that the faster you go, the higher the accelerations and the more unstable the system and the brute force stuff just keeps on looking better and better. jerry
|