Hello,

I had a very good time in Pisa at the hackmeeting with the Netsukuku
team! That was already 22 days ago! I want to apologize for not
writing before.

In Pisa, me and some other people from University of Rome Tor Vergata,
had the opportunity to talk with AlpT to make questions regarding the
QSPN

To our feeling after the hackmeeting is that there are very good ideas
behind the project, but somehow this ideas should be written in a
proper way.

The excellent work carried out by AlpT in the documentation, only
explains the theory behind the netsukuku network protocol.

We strongly believe that writing a specification, in parallel during
the implementation, can prevent protocol design problems that might
arise in the future.


First problem.

As far as we understood a Netsukuku node to execute the hook procedure
has to choose a random IP address, and exchange maps with neighbours
nodes.
We strongly believe this is unnecessary (choose the temporary IP) and
should be managed in UDP.
You might want to read Noa-OLSR and have a look at their approach.

Second problem.

This is about metrics.
As far as our knowdge IP protocol  uses hop count as metric. Problems
started to arise when the TCP/IP protocol stack moved from on top of
wired connections to wireless connections.
IP itself does not provide the necessary tools to integrate new
metrics (protocol design problem) so we adjust the hop count (faking
more hops than real) in order to cope with.
I mean that in a routing table we have an integer value that should
represent the hop count, and that's the metric to choose the best
route. Some protocols use that valure related to another metric
different than hop count.
As an example OLSR or batman feed the routing table of the kernel with
hops calculated by other means than hop count.
So all this was just to say that IP is not very friendly in adding a
new kind of metric.

The biggest NEW problem comes taking into consideration asymmetric
links (that in the wired world didn't exist if we don't think about
ADSL lines that are asymmetric for choice of ISP).

Now consider two nodes A and B phisically connected with a wireless
link. In netsukuku a TP Generated by the node A and received by the
node B  let's be node that there is a link A->B but B can't be sure
that there is also a B->A node, and there is an high probability that
bandwidth of A->B is very different than bandwidth of B->A.

So, if you are writing Netsukuku in order to be adaptable to many
metrics, you should make it adaptable also di metrics that take into
consideration asymmetric links. This means signalling of QSPN should
be able to assign 2 metric values per link.

my 2 cents

feel free to go ahead with questions :)

Saverio Proto (and some more Tor Vergata people)
_______________________________________________
Netsukuku mailing list
[email protected]
http://lists.dyne.org/mailman/listinfo/netsukuku

Reply via email to