[Olsr-dev] Meshing over VPNs
Teco Boot
(spam-protected)
Mon Oct 10 18:44:17 CEST 2011
Op 10 okt 2011, om 11:20 heeft Markus Kittenberger het volgende geschreven:
> (but i aim for scalability to thousands of vpn peers)
+1
I use this model:
- I use OpenVPN, with tunnel to a VPN "relay server"
on each olsr router
- I don't run OLSR or any other routing protocol on
the VPN tunnels
- I have a mapping between certificate "common_name"
and HNA route, for each router
- After tunnel setup, bidirectional routes are set up:
o Uplink: all RFC1918 routes (10/8,172.16/12 and
192.168/16) into tunnel
o Downlink: the HNA route for that router, generated
from common_name (in IP and VPN routing tables)
- VPN servers inject these HNA routes into a backbone
routing protocol (in my case: OSPF). This provides
connectivity to central servers and other MANET nodes,
in different clusters (no direct OLSR paths).
This results in no overhead in routing on tunnels. The *only*
overhead is tunnel keep-alive traffic, and app keep-alives if
used. I see no problems in scaling up to 10^4 - 10^6 nodes,
even on constraint uplinks (which I have: from broadband
Internet access to 3G/2G/satcom.
This model does not support reach-back to nodes behind an
OLSR router that doesn't run OpenVPN. This would require
running some form of a routing protocol on the VPN tunnel.
I played with BGP and OSPF, together with some route
distribution. But if massive scalability and massive
multihoming is needed, the described model fits far better.
As a next step, I think of an improved SmartGateway, with
VPN tunnel load balancing.
Improved SmartGateway:
- set up new SmartGateway tunnels if a (much) better
gateway is detected, and use this for new connections
- tear down inferior / unused SmartGateway tunnels
(this could break connections!!)
VPN tunnel load balancing:
- Use MPTCP for OpenVPN (oeps, TCP instead of UDP)
- synchronize MPTCP with SmartGateway
Opinions?
Thanks, Teco
More information about the Olsr-dev
mailing list