On Sat, Oct 2, 2010 at 4:05 PM, ZioPRoTo (Saverio Proto)
<[email protected]> wrote:
>> Right now netsukuku does not use tinc, but very soon [my experimantal
>> branch] will do.
>> Pairs of nodes will (optionally) use it to raise the internal
>> connectivity of a gnode, hence lowering the probability that some
>> nodes could need to change their IP.
>
> if your goal is to avoid network addresses collisions, please review
> the literature to solve that spefic problem.

The goal is not that. It is to avoid that a gnode becomes internally
not connected.
See chapter 7.2 of topology.pdf in documentation, second case.
When a gnode becomes broken (not internally connected) a part of the
nodes have to hook again to the network obtaining a different address.
This is due because of the hierarchical topology in use in netsukuku.
The use of tinc permits to create virtual links between border nodes
of a same gnode. This will not eliminate the problem, but will
mitigate its chance of happening.
The overhead added by radar traffic should not be huge. If the split
does not happen then the virtual link has little chance to be used by
normal traffic because it has a very high round trip time.
Should a split happen, then the overhead in normal traffic will occur
but this will avoid that a node changes IP, so it is better than
worse. A badly performing link should in any case induce a node to
leave a gnode in order to find a better situation in another gnode
(see chapter 7.3), so the overhead will be temporary anyway.

Apart from these hypothesis, the code is experimental, we never had
tests to measure how good this behaves.

--Luca
_______________________________________________
Netsukuku mailing list
[email protected]
http://lists.dyne.org/mailman/listinfo/netsukuku

Reply via email to