[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [manet] Default metric in OLSRv2



Hello, Henning,

I would add two points to the discussion which is interesting and important.

First point: there are many ways to have a MANET/OLSR not working well: bugs, badly tuned parameters. The main source of trouble is in general how neighbor sensing asserts the quality of a link in order to accept it or reject it, and I agree that using a metric will greatly help.

Anyhow RFC 3626 was not excluding the use of link quality in neighbor sensing, and was suggesting some. The hop count should not be problem as long as the links that are on the path are good links. In other words I would see the use of path quality as a minor improvement over the use link quality in neighbor sensing. Anyhow this would help traffic engineering.

Second point: adding a default link quality in OLSRv2 may be as complicated as inserting general metric in OLSv2, eg in MPR selection, so we might be careful if we hurry too much.

Anyhow I am looking forward a draft that describes the ETX metric that has got very good experimental records and personnaly I like the dimensionless metrics (we designed and experimented some with IEEE 802.11/ OLSR cross layering).

Best regards


Henning Rogge <hrogge at googlemail.com> a ?it :

Hello,

I know this mail comes a little bit early (metrics have not been merged into
OLSRv2 draft), but I would like to like to argue for more than "include
generic metric support in OLSRv2".

Two weeks ago (just before the OLSR Interop 2009 in Vienna) I visited a
conference about military communications. There were a couple of presentations
which used OLSR as a mesh net protocol and they can be summarized as: "OLSR
does not work well". When I tried to get some more facts about this I quickly
recognized that all of them used hopcount metrics "because it's the default
described in the RFC".

I think this was a mistake in the OLSRv1 RFC, it described a protocol that did not work very well in the real world because of hopcount metrics. No amount of
great research for new metric or improved features have changed this later,
because most people USING OLSR for their research will use the stuff described
in the default RFC. We should not do the same mistake in the OLSRv2 RFC.

My suggestion is:
include a simple version of ETX as an example metric into the OLSRv2 RFC.

Why ETX ? Why not ETT/SNR/MIC/.... ?

ETX is not the best metric out there, it's a very basic one. But it's VERY
easy to describe and implement, it's independant from the layer 1/2 below OLSR
and it allows to build much better real world networks than hopcount. I know
no other metric that can provide these three advantages.

We should add a few sentences into the chapter that this will be just the
"lowest common metric" which should work with all OLSRv2 agents and that there
are much better and more specialized metrics described in other documents and
research papers. But if we do not put a real metric into the RFC we will have
people saying "OLSRv2 does not work well" in years, just because they do their
comparision based on the RFC implementation (hopcount).

If there is a consensus on this I can easily write description of a simple
Hello-Message based ETX implementation with a single tuning parameter for the
draft.

What do you think ?

Henning Rogge




--
Philippe Jacquet
Hipercom project team leader
INRIA
tel: +33 1 39 63 52 63
fax: +33 1 39 63 55 66

----------------------------------------------------------------
This message was sent using IMP, the Internet Messaging Program.