[RFC] Connection oriented routing

[RFC] Connection oriented routing

[Michael Blizek]

Hi!

I am programming a new layer 3(+4) protocol for zero administration
(community/mesh) networks. It is source routed. Routers know about the
connections and can optionally do onion like reencryption. There is also a
QoS/credit system with explicit Xon/Xoff instead of "congestion avoidance". A
few things are already running (see
http://michaelblizek.twilightparadox.com/projects/cor/index.html ) for the
code and a few more ressources.

The drawback will be that ressource usage on the routers is likely higher than
IP. At the moment I do not know exactly how much higher. It seems like the
price, I just hope it is not too high.

	-Michi

[RFC] Connection oriented routing

[Michael Blizek]

Hi!

I am implementing a connection oriented source routed layer 3/4 protocol for
zero administration (community/mesh) networks. It takes care of things like
fair bandwidth allocation and has higher security/privacy so that it can really
run without administration. It is basically running, but not entirely stable
and with some loose ends. You can find some more info here:
http://michaelblizek.twilightparadox.com/projects/cor/index.html

git repo (based on 2.6.28):
http://repo.or.cz/w/cor.git?a=tree

My 2 previous queries to netdev:
http://article.gmane.org/gmane.linux.network/80232
http://article.gmane.org/gmane.linux.network/144915

	-Michi

[RFC] Connection oriented routing

[michi1]

Hi!

I have been deleloping a connection oriented layer 3+4 protocol mostly for
wifi mesh networks. Its main difference is that "tcp" connections are not
end-to-end but between neighboring routers instead.

It runs ane you can test it, if you want, but it is not ready for merging or
actual use yet. Some unfinished tasks include:
- cleanups
- performance: lots of things which can be improved
- keepalive
- interfaces for configuration+statistics
- move to a newer upstream kernel
- I am not sure how some of the in-kernel interfaces work. More specifically,
  I am not sure which "struct proto/proto_ops" functions need to be
  implemented and how ho avoid race conditions when accessing various
  variables. Also, network namespaces are not handled yet either. I will look
  closer into this when the other issues are resolved in order to avoid
  upstream changes breaking it in the mean time.

Website: https://michaelblizek.twilightparadox.com/projects/cor/index.html
Kernel repo: https://repo.or.cz/w/cor.git?a=tree
Userspace tools: https://repo.or.cz/w/corutils.git

Please share your thoughts!

	-Michi