[Olsr-dev] Patch Orgy Syncup

Juliusz Chroboczek (spam-protected)
Thu Aug 23 23:58:41 CEST 2007


I hope this discussion is okay for this list.  At any rate, I need to
mention that I've spent much of the last six months designing
a distance-vector protocol for mesh networks[1], so I'm probably as much
biased in favour of DV as you are in favour of LS.

>  BGP with iBGP speakers preturbing the distributed computation 
>    because of damping (read: bogus transient routes)
>
>  and
>
>  OLSR not installing its routes (in order to serve the hopbyhop
>  routing paradigm) is semantically similar.
>
> either way the net results is the same - routing and forwarding loops.

Yes, but the reasons are very different.

Distance Vector is intrinsically robust in the face of inconsistent
routing decisions.  As long as every router only advertises routes
that it is actually using, and with a metric that is no less than the
metric that it's actually using, DV will converge, possibly on
a suboptimal set of routes.  Link State, on the other hand, fails to
converge on a loop-free forwarding graph unless all the routers make
the same shortest path calculations.

That is the fundamental reason why you can do arbitrary aggregation,
filtering and policy routing in DV, but not in LS.  (To be fair, OSPF
allows some aggregation and filtering to happen at area boundaries.)

(Distance Vector has other problems, of course, most notably the fact
that unless you're careful convergence to a loop-free graph can take
a very long time; the so-called ``counting to infinity'' problem.)

The issue with BGP is different.  BGP does not use hop count as
a metric for computing shortest routes; very roughly speaking, it
minimises the number of ASes crossed by a route.  Hence, if two routes
are advertised by routers in the same AS, BGP considers them as
equivalent; there is no mechanism to avoid loops within an AS.  Hence,
BGP assumes consistent routing information within an AS, and requires
careful synchronisation with the internal routing protocol.

I am not aware of any such issues with DV protocols that use hop count
(or a metric that increases with hop count, as when you measure link
quality).

                                        Juliusz




More information about the Olsr-dev mailing list