Re: [manet] Default metric in OLSRv2

["Dearlove, Christopher (UK)" <Chris.Dearlove at baesystems.com>]

> My observation in one
> scenario is a source node uses a forward path that is different from a
> return path from a destination node, both paths 3 hops apart. So my
question
> is does this commonly happen using OLSR with hop count as a metric?

Yes.

Let's consider the simplest interesting network, one with four nodes
in a square, with no diagonal connections, ABCD round the square.
I'm going to discuss two hop paths, obviously this generalises to
more than two hops.

Actually any node can send information "left" or "right" to its
opposite node, because it can use two hop information directly
without considering MPRs. But use of MPRs is preferred by RFC
3626 (and in OLSRv2) so we'll assume that.

What that means is that if A is routing to C it will choose
B as a preferred route if it has a B to C link in its Topology Set,
i.e. (assuming only MPR selectors are advertised in TC messages)
if C selected B as an MPR. A will choose D as a preferred route
if C selected D as an MPR. C must choose one or the other. It may
choose both if there is more to this network, in which case A has
a choice, rather than its choice being dictated by C.

But in general that means choices. Choices can't be avoided (with
extra nodes hanging off B and D, C must choose both as MPRs).
Sometimes those choices are made (in this case) by A, sometimes
by C, sometiimes (in more complicated cases) by A and C and by
other nodes on the path between them.

Going back to the square, we can in that case make consistent
choices (other than by extra signalling, which rather defeats
the object of the OLSR exercise) only by consistent MPR
selection. In general that's not possible. In the square case
it's not impossible, some coordination can be done. But without
coordination everyone might pick MPRs such that all traffic
flow over two hops is, say, clockwise.

There's also a question as to whether this matters, and if so
why. But if that is a desirable feature then you need either
a significant modification of OLSR (and not having considered
it I don't know how easy it would be, but my guess is that it
would be tricky with limited topology, easier with fuller
topology) or a new routing algorithm. (And how maintainable
it is in the face of dynamism would be another question.)

And finally, if a wide range of metrics is used, and metrics
are the same in both directions, then this gets less likely
with metrics (as it requires what becomes more of a coincidence
of there being equal length routes). But with unidirectional
metrics (as we intend to add) it becomes quite likely with
metrics also.



********************************************************************
This email and any attachments are confidential to the intended
recipient and may also be privileged. If you are not the intended
recipient please delete it from your system and notify the sender.
You should not copy it or use it for any purpose nor disclose or
distribute its contents to any other person.
********************************************************************