[Olsr-dev] reliable vs unreliable updates

[Juliusz Chroboczek]

> Reliable updates could be useful for DB synchronizing after a startup
> or manet merge. Fist goal would be fast DB exchange, either reliable
> or not.

As you justly note, that's pretty much unrelated to the issue of
reliability.

FWIW, Babel does a fast resynchronisation at startup and under some
circumstances that are indicative a merge[1], although it does so
unreliably.  This has been shown to be beneficial in practice: it allows
a node to become productive just one or two seconds after it finishes
booting[1].  But there's no evidence whether doing that reliably would
be beneficial (just to be clear: there's no evidence either way).

                                        Juliusz

[1] babeld.c line 515 and neighbour.c line 187.

[2] That's because babel also sends a gratuitious hello when it sees
    a new node, which accelerates ETX convergence.  (And before you
    complain, Babel can perform accurate hello counts even in the
    presence of gratuitious hellos.)