Re: [manet] CDS robustness research repport

[Richard Ogier <ogier at earthlink.net>]

Philippe,

Philippe Jacquet wrote:

> Hello, Richard,
>
> Thank you for your comment and results. You made very interesting 
> remarks. It was already pointed out by Dai and Wu that neighbor 
> designated algorithms are more robust.
>
> MPR based CDS uses only two hop information as well as MPR selection 
> in OLSR, no need of three hop information.

This is often misunderstood, so I will explain more clearly.
You need to specify what you mean by "2-hop information",
i.e., 2-hop information with respect to which node?
Each node selects its MPRs based on 2-hop information (with respect
to itself). But (with offline pruning) each node decides whether it
is a CDS node based on the MPR selection of its neighbors, which in
turn depends on the 2-hop information for each neighbor.
Thus, the decision of whether a given node becomes a CDS node
depends on 3-hop information with respect to the node itself.
This is the tradeoff, so it is important to understand this.
Do you agree?  If not, then please explain.

BTW, I wanted to add something to my previous message.
The MDR selection algorithm in my draft selects MDRs persistently,
which is important for the stability of adjacencies.
In order to modify the OSPF-MDR extension to use an MPR-based
CDS, it would be necessary to use a *persistent* MPR selection
algorithm (and also for selecting Backup MPRs).  This will likely
result in a larger number of MPRs and and larger CDS than shown
in your simulations (since you used a non-persistent MPR selection
algorithm).

Another point is that it is not clear how much more robust an MPR-based
CDS will be if reliable flooding (using adjacencies) is used as in
the OSPF-MDR extension. That is why detailed simulations are necessary,
e.g., using GTNetS.

Regards,
Richard


> It is  a little less efficient than MPR flooding but as resilient and 
> does not need last hop identification for broadcasting data, this is 
> why it is in our SMURF draft for data broadcast.
>
> "Rule k" algorithm is originally from Dai and Wu, this is a 
> generalization of previous rule k=1,2,3 algorithms of Wu Li that are 
> very heavy on computation, k=infinity is more efficient and easier to 
> implement. It also needs two hop information as with MPR. I think your 
> MDR algorithm is a kind of adaptation of rule k=infinity.
>
> I was not aware of the stretch factor effect, this is an interesting 
> remark.
>
> Best regards,
> Philippe



_______________________________________________
manet mailing list
manet at ietf.org
https://www1.ietf.org/mailman/listinfo/manet