thanks much for you comments! indeed, topology dynamics is concern #1 in geometric routing. that's why we considered both short- and long-term dynamics, all in the paper. we emulate the former (by killing a percentage of links and nodes) and replayed the latter using the measurable history of internet evolution over the past few years with ASs and AS connections appearing, disappearing, etc., and the results are still very good, pretty much the same as for the static case. amazing, isn't it? i know it's hard to believe, and even we can't stop being surprised how well it works. we have another paper in submission, where we take space to explain why it works so well, and where we discuss some aspects of what it would take to implement and use this stuff in practice. -- dima. http://www.caida.org/~dima/
On Friday, September 10, 2010 4:29 PM, Sampo Syreeni wrote: > On 2010-09-09, Dmitri Krioukov wrote: > >> marshall, thanks for posting it here. i also thinks it's relevant :) > > Thanks from me too, and it's certainly relevant. Still, it might not be > as good an idea as it sells itself as. > > Geometric routing ideas have been around for quite a while now. They > certainly do this sort of thing within manets right now, because of the > spatial nature of a cloud of terminals/sensors. So in certain ways the > idea works well indeed. > > I'd be the first to say that geometric routing is a swell and elegant > idea. Yet, it tends to have some inherent problems in the wired setting > where a) the topology and the geometry of the network isn't as static as > a cloud of 3D sensors would see, b) where we have to have static contact > points like DNS fully available at more or less fixed destination > addresses all of the time, to map from points of interest to > topological/geometrical addresses/locations, c) any static mapping like > the one proposed in the paper could *severely* undercut routing > efficiency as soon as someboby built a new undersea cable, which of > course severely changes the routing landscape in one fell swoop, and d) > when we then probably would go with an adaptive routing protocol, there > is a serious problem with asymmetric paths. That final problem doesn't > plague just Euclidean distance measures, but all of the metric ones as > well, including the hyperbolic. > > As regards an adaptive geometric routing protocol, IRTF's ALTO group has > charted this stuff quite extensively already in the context of routing > within overlay networks. I suggest everybody look into that body if they > haven't already, if interested in geometric routing. > > In my opinion, this particular article is a nice touch onto how best > parametrize network distance. Based on the article and the references, a > hyperbolic space might well provide us with a better parametrization of > distance in a scale-free network within the geometric routing paradigm. > But it won't solve the more fundamental problems which have stopped us > from adopting geometric routing in the past. > > I'd say this body of work is a building block for further research, more > than the showstopper it'd like us to see itself as. > -- > Sampo Syreeni, aka decoy - [email protected], http://decoy.iki.fi/front > +358-50-5756111, 025E D175 ABE5 027C 9494 EEB0 E090 8BA9 0509 85C2 _______________________________________________ rrg mailing list [email protected] http://www.irtf.org/mailman/listinfo/rrg
