Thank you, a reply so full of insights! Please let us know what you decide -- I'll be glad to give a hand if > you're not sure what to do.
mm, up to now we imposed every node to originate at least a route, so that it was easy to disseminate router-ids in the whole network and run the centrality algorithm. But to be more general and if we really want to keep thinking at a node-layer, then it seems to be no escape: we will have to implement a per-node flooding mechanism... I am wondering if we can be more elegant and integrate our algo with Babel relying only on interface-identifiers or prefixes, somehow avoiding flooding, or even if it is worth to stop thinking at a node-layer and consider the network at a "prefix-layer"... So well, we definitely do not have any decision for the moment, we need some time to think about it in much more details :) Thank so much for now, Lorenzo 2018-01-25 16:23 GMT+01:00 Juliusz Chroboczek <j...@irif.fr>: > Ok, I see. > > > The algorithm let every node in a network compute its own value of "Load > > Centrality" and afterwards let nodes disseminate computed indexes so > > that each node, at convergence, is aware of the centrality index of all > > nodes in the network. > > [...] > > > For the moment forget that actually Updates advertise distances > > associated to prefixes, not to router-ids...we just wanted to work on > > a network graph-layer. > > That's the whole point. Babel is a distance vector protocol, it > advertises routes to given prefixes. Now, of corse, prefixes are > originated by nodes (that's what the router-id identifies), but Babel does > not flood information about nodes: > > - if a node originates no route, then it doesn't appear on the wire at > all (Hello and IHUs carry interface identifiers, not node identifiers); > - if a given prefix is advertised by multiple routers, then parts of the > network will only see one of the routes to that prefix; > - finally, Babel doesn't use a reliable transport -- under heavy packet > loss, part of the information may be dropped (which is okay for its > intended purpose -- if there's such heavy packet loss, then there's no > point advertising routing information, since the routes are unusable > anyway). > > In other words, Babel, as it currently stands, does not have provisions > for advertising node-specific information globally. > > > Do you think the "transitive attributes" feature could be of general > interest? > > As a general rule, we try to avoid adding to Babel features that are not > useful for routing. You could do one of the following: > > - design your own reliable flooding protocol; that's not difficult, I've > given it as a project to fourth year students and almost everyone > succeeded; > - piggyback on a protocol that does reliable flooding, such as IS-IS or > OSPF; > - piggyback on a protocol that does unreliable flooding, such as OLSR or > HNCP; > - add a subprotocol for flooding of per-node information to Babel, but > to be entirely frank with you, I doubt it will get into mainline > babeld. > > Please let us know what you decide -- I'll be glad to give a hand if > you're not sure what to do. > > -- Juliusz >
_______________________________________________ Babel-users mailing list Babel-users@lists.alioth.debian.org http://lists.alioth.debian.org/cgi-bin/mailman/listinfo/babel-users