[B.A.T.M.A.N.] path selection

[B.A.T.M.A.N.] path selection

[axel]

[-- Attachment #1: Type: text/plain, Size: 1564 bytes --]

Hi,
there is also one concern bothering my mind since a while.

I think the protocol currently establishes routes between nodes, but optimizes 
them the wrong way around. More concrete, the originatorMessages (OGMs) 
initiated by each node install routing information for DOWN-link traffic in 
the mesh which are actually optimized for UP-link traffic. 

The following scenario and attached figure (is of course a little bit 
constructed but it) may illustrate this. The figure shows 4 nodes (A,b,c,D) 
and 4 existing links between them A-b, b-D, A-c, and c-D - thus, two 
potential routes between node A and D. The  links b-D and c-D are symmetric, 
perfect links with 0% packet loss. The links A-b and A-c are asymmetric links 
with 0% packet loss for A->b and A<-c, but 50% loss for A<-b and A->c (See 
attached figure). However, each of the 4 links can be assumed as a 
bidirectional link since at least every second OGM will reach the 
corresponding link neighbor. 
Now, node A would receive 100% of the OGMs initiated by D and rebroadcasted 
via node c but wouldreceive only 50% via node b. Therefore A would select c 
as its best nighbor towards D (obversely D would select b as its best 
neighbortowards A). 
However, in this case, that is not the best choice since every second packet 
send via A-c-D needs to be retransmitted on the link A-c, which would not be 
necessary if send via A-b-D . 

Don't know if you agree, ...is that reasonable? Also I don't have any simple 
approach in mind to solve this but it might be worth to reconsider.

ciao,
axel


[-- Attachment #2: asymmetricPathChaos.jpg --]
[-- Type: image/jpeg, Size: 6802 bytes --]

Re: [B.A.T.M.A.N.] path selection

[Marek Lindner]

Hi,

> However, in this case, that is not the best choice since every second
> packet send via A-c-D needs to be retransmitted on the link A-c, which
> would not be necessary if send via A-b-D .
> Don't know if you agree, ...is that reasonable? Also I don't have any
> simple approach in mind to solve this but it might be worth to reconsider.

I think you made a good point but I even would go a step further. Consider 
this: A downloads a file from D thus using D-b-A  (with 50% packet loss). But 
each packet has to be acknowledged and every second acknowledgement is also 
lost (via A-c-D) . That is why D will resend every second packet and the 
throughput will suffer even more.

I think we should find a solution here but up to now I don't have a clue how 
to do it.

Regards,
Marek