Wed Feb 9 14:27:38 CET 2005
> 1. Let's assume that we look at a path from node A to node B via
> intermediate nodes and that we want to know the total metric of the path
> in order to minimize it to find the best path. Then there are three
> different kinds of metrics:
there you are going into the field of true QoS, which leaves the area of
routing with tables. you will have to decide on a per-packet basis,
which route to take, since each packets defines his own contraints
(delay, bandwidth, jitter).
some ideas how this could look are at
even worse, to obtain the best route in a multiplicative metric
environment, dijkstras algorithm does not work any longer. according to
my graph algorthmics professor (prof. mutzel) you will have to resort to
an algorithm that has a O(2^n) runtime :-( or come up with a heuristic.
> implementation minimizes additive edge weights. The question now is
> whether there is a way of converting multiplicative metrics and minimum
> metrics into additive metrics in a meaningful way.
i don't think so.
> I think that the biggest problem is that you need a pretty big testbed
> to test any modification to the employed metric. However, each
> modification has the potential of completely breaking an existing mesh,
> if the new metric turns out to be unsuitable.
some parts of our mesh are also routed statically, in a different
subnet, which causes great pain in various body parts to configure.
delpoying and later removing a potentially unstable olsrd would be possible.
More information about the Olsr-users