[EMAIL PROTECTED] a écrit : > Send Babel-users mailing list submissions to > [email protected] > > To subscribe or unsubscribe via the World Wide Web, visit > http://lists.alioth.debian.org/mailman/listinfo/babel-users > or, via email, send a message with subject or body 'help' to > [EMAIL PROTECTED] > > You can reach the person managing the list at > [EMAIL PROTECTED] > > When replying, please edit your Subject line so it is more specific > than "Re: Contents of Babel-users digest..." > > ------------------------------------------------------------------------ > > Today's Topics: > > 1. Fixes to route selection (Juliusz Chroboczek) > 2. ANNOUNCE: Babel 0.10 (Juliusz Chroboczek) > > > ------------------------------------------------------------------------ > > Sujet: > [Babel-users] Fixes to route selection > Expéditeur: > Juliusz Chroboczek <[EMAIL PROTECTED]> > Date: > Tue, 11 Mar 2008 19:01:39 +0100 > Destinataire: > [email protected] > > Destinataire: > [email protected] > > Precedence: > list > Version de MIME: > 1.0 > ID du Message: > <[EMAIL PROTECTED]> > Content-Type: > text/plain > Message: > 1 > > > Dear all, > > Lilian has found a case in which Babel 0.9 would delay reconvergence > by 30 seconds. I've now fixed this case, and reworked the route > selection algorithm somewhat; convergence speed is much improved. > > Since I'd like to release 0.10 pretty soon, I'm very much interested > in any data about the behaviour of the current Darcs tree that you may > have. > > Since I'm in a chatty mood, I'll give a few details about the issue. > (By the way, Lilian, my analysis of the problem on IRC was completely > wrong.) > > A (proactive) routing protocol has three tasks: > > - neighbour sensing: measuring link quality between neighbours; > - route discovery: discovering the set of available routes; > - route selection: choosing a particular route to every destination. > > In many routing protocols, the three tasks are mixed up. For example, > both RIP and B.A.T.M.A.N. mix up neighbour sensing and route discovery, > which is why in RIP you cannot tune the hello interval separately from > the updte interval. > > Babel was designed as a modular protocol: the three stages of the > protocol are, to the extent possible, independent. Of particular > interest is the independence of route selection and route discovery: > the protocol remains correct if nodes do not always choose the route > with smallest metric, but choose ``worse'' routes under some > circumstances (as is usually the case with distance-vector protocols, > but not with link-state protocols). This makes it possible to > implement route selection policies that choose routes according to > criteria richer than just link quality. > > Now, when I started working on Babel, I had just experimented with > OLSR, and found that it has serious problems with route stability: > under some circumstances, it flaps between two parallel routes to > a given destination. Route flapping is ungood: it can cause packet > reordering, and it usually causes jitter (for example due to an extra > ARP exchange) and sometimes packet loss. > > Because of that, Babel's route selection policy tries to promote route > stability. The main stability mechanism is a simple form of > hysteresis: we don't switch routes unless the new route is > significantly better than the old route. (The exact amount of > hysteresis used depends on whether we also switch router-id, and is > something that needs more experimentation.) > > The second route stability mechanism is much simpler: a route is > regarded with great suspicion unless it has been around for 30 > seconds. Unfortunately, this mechanism turned out to be buggy, and to > interact badly with some other features of the protocol (notably the > feasibility condition), which caused perfectly good routes to be > disregarded for 30 seconds in Lilian's network topology. > > After some thought, I've simply removed the 30 second condition; the > only route stability mechanism remaining is hysteresis, which appears > to do The Right Thing in practice. > > In case you want to check the code, all of this happens in route.c, in > the functions find_best_route (which finds the candidate route for > a given destination) and consider_route (which decides whether to > switch to a new, potentially ``better'' route). > > While I was at it, I reworked some of the constants in the protocol, > notably by making it send multi-hop requests and triggered updates > more aggressively; this makes the protocol somewhat more chatty, but > very slightly so. All together, you should see much faster convergence > with 0.10 than with 0.9. > > Juliusz > > > > > ------------------------------------------------------------------------ > > Sujet: > [Babel-users] ANNOUNCE: Babel 0.10 > Expéditeur: > Juliusz Chroboczek <[EMAIL PROTECTED]> > Date: > Wed, 12 Mar 2008 00:52:31 +0100 > Destinataire: > [email protected] > > Destinataire: > [email protected] > > Precedence: > list > Version de MIME: > 1.0 > ID du Message: > <[EMAIL PROTECTED]> > Content-Type: > multipart/signed; boundary="=-=-="; micalg=pgp-sha1; > protocol="application/pgp-signature" > Message: > 2 > > > Dear all, > > I am pleased to announce Babel 0.10, which you will find on > > http://www.pps.jussieu.fr/~jch/software/files/babel-0.10.tar.gz > http://www.pps.jussieu.fr/~jch/software/files/babel-0.10.tar.gz.asc > > This version will automatically redistribute local addresses; it is no > longer necessary to specify them using ``-X''. This is an incompatible > change, but you can revert to the previous behaviour by specifying > > redistribute local deny > > in your /etc/babel.conf. > > This version will also automatically adapt to interfaces appearing and > disappearing at runtime, which avoids the need to restart Babel when > a VPN goes up or down. > > Some fairly significant improvements have been done to the speed with > which Babel selects a new route when a neighbour goes down (as opposed > to the case where the local node moves). Many thanks to Lilian for > helping me debug that. > > Finally, Babel can now daemonise itself, which should simplify your > init scripts. > > I strongly recommend upgrading from Babel 0.9 or earlier. > > Juliusz > > 11 March 2008: babel 0.10 > > * Implemented the ability to automatically export local addresses (see > the ``local'' keyword in redistribute specifications). This should > avoid the need to explicitly specify -X on the command line > (Julien Cristau and Juliusz Chroboczek). > * INCOMPATIBLE CHANGE: local routes (local interface addresses) are > now exported by default. Specify ``redistribute local deny'' to > avoid that. > * Babel will now automatically choose a router id if none is > specified on the command line. > * Automatically adapt to interfaces appearing or disappearing at runtime, > as is usually the case when running over tunnels or VPNs. > * Changed the link quality computation algorithm to not discard very > lossy links. > * Multi-hop requests will now be forwarded to an unfeasible successor > under some circumstances. > * Send multi-hop requests more aggressively. > * Send requests for a new seqno upon receiving an unfeasible update > if it's better than what we have. > * No longer consider the age of routes in route selection. > * Added ability to run as a daemon. > > ------------------------------------------------------------------------ > > _______________________________________________ > Babel-users mailing list > [email protected] > http://lists.alioth.debian.org/mailman/listinfo/babel-users > Hello,
I submitted a ticket to update babel in openwrt https://dev.openwrt.org/ticket/3230 Sincerely, Lilian _______________________________________________ Babel-users mailing list [email protected] http://lists.alioth.debian.org/mailman/listinfo/babel-users
