Well, as soon as something is changed in the protocol
you lose the big advantage of
running standard protocols available in standard firmware, see the AMPR RIP thing.
Not changing the protocol, but finding the right protocol that can
accomodate with link quality and weight.
I have considered doing similar things in BGP
(adjusting prepend or local-pref
dynamically based on SNMP monitoring of the link).
We have not experimented with OSPF yet, I read in many places that it has problems
with scaling when the CPU power on routers is limited (like on old RB750s)
I don't know Mikrotik. We are using OpenWRT, mostly on TP-Link or
GL-Inet cheap routers. The border routers are Debian VMs, with
potentially unlimited CPU. And my OSPF network does not have a size
which can disturb a CPU, even on a $20 hardware :-)
This is an interesing purpose, anyway. Maybe it's better to keep two
separate routing policies :
- one protocol for external/internet, which should be BGP
- another protocol for internal routing, which can be GBP, or something
else
But this only applies for networks build around a regional/local
gateway. For a standalone endpoint (such as current IP-IP user), there's
no distinction between "internal" and "external" routing, there's
only
one routing policy.
Then, the question becomes :
- Is it better to keep full mesh / standalone endpoints (such as current
IP-IP) ? But if so, how to handle Plug and Play and NAT traversal ?
- Or is it better to have small local gateways managed by skilled teams,
and end-users connecting to those gateways with simpler PnP VPN systems ?
We choosed the second option, with fully home-made design (OpenWRT,
OpenVPN, OSPF), because it best suited our needs, and because we are an
island, with few inter-connects with the rest of the world.
It seems lots of people in the world are using similar designs, with a
central gateway and enpoints connecting to it via VPNs. Maybe we just
have to share our experiences, and adopt some kind of "standardized"
rules for our gateways ?
73 de TK1BI