Re: [manet] Default metric in OLSRv2

["Teco Boot" <teco at inf-net.nl>]

I support a strong recommendation for using link metrics in a MANET.

I think ETX is the poor men solution, which I would not recommend
for real networks.
But yes, ETX is not dependent of sub-IP layer information.
I've no objection against adding it in an appendix.

Reading dearlove-olsrv2-metrics-04, it is recommended that in a 
well-designed network, all routers will use the same physical
metric type. I agree on this, but I highly prefer an OLSR specification
that uses dimensionless metrics, in TC and Hello. This enables using
different physical metric types, without incompatibility. 
I'm not sure if this is fully supported in dearlove-olsrv2-metrics-04.
Using for example ETX and sub-IP layer information such as RFC4938 in
the same network and a need of transfer of data from receiver to sender
(e.g. RSSI, hello loss ratio etc.) could cause problems. Let's try to
avoid such problems.
My suggestion: When router A is the originator of a TC, it is 
responsible for the advertized metric for outbound, from A to whatever
neighbor. A's neighbors should send a helpful, dimensionless metric to A.
In addition, the neighbors may send metrics to A with dimensions, 
but MUST NOT rely on these metrics being used.

A reason for neighbors of A not sending a helpful, dimensionless metric 
to A is the awareness of A having the information already (e.g. the sub-IP
layer transfers such information already) or too much overhead.

Regards, Teco