From: Lillie Kittredge (kittredl@u.washington.edu)
Date: Sun Nov 21 2004 - 23:03:45 PST
This paper discusses a location-based routing protocol for wireless
networks.
Greedy perimeter stateless routing uses graph theory to totally show other
routing protocols who's boss. The key theoretical insight is to see the
collection of wireless nodes as a planar graph, and to prune unnecessary
edges using methods that can make decisions locally; decisions that each
node can make knowing only its neighbor's locations. After a node has
determined which of its neighbors it should ever send to, it uses the
available information to rout packets towards their final physical
destination. Two modes deal with changing topology as the packet
traverses the graph: greedy routing early on, and a perimeter routing to
get around the areas where greedy routing won't work.
Speaking as someone working with ns currently, I am impressed at their
restraint in discussing it. The issue of extra functionality in the dsr
implementation is exactly the kind of thing that crops up all over in what
is, especially in its wireless model, a frustrating kluge. I note their
concern about the difference between the published results of dsr
simulation and their own results: I would not be surprised if something
undocumented had changed in ns between Broch's simulation and theirs.
The major thing that I found unsatisfying about GPSR is that the source
must know the location of the destination node when sending the packet.
They gloss over the details of the name->location lookup server, but my
concern is how current the information will be by the time the packet gets
there. They admit that addition of a dns-like service to map names to
locations will add overhead, but I wish they'd address the issue that
nodes may move and interfere.
The other issue I find disconcerting is the assumption of nodes all at the
same level. Though for many applications this is acceptable, it is likely
to cause intense frustration later on. A sensor network describing one
floor of a building is all well and good, but why not one describing the
whole 3 dimensional duct system? For sensor networks in natural
environments, also, it is verging on rather a lot to ask that they remain
more or less coplanar.
This archive was generated by hypermail 2.1.6 : Sun Nov 21 2004 - 23:03:46 PST