Sat Apr 24, 2021 3:33 am
@millenium7, OSPFv3 only drops the specific adjacency for the interface changed. Unlike OSPFv2 that freaks out entirely, this is pretty graceful. The mesh would be pretty static and stable so if I'm intending to reduce an interface cost because of poor modulation it's probably ok to drop this. would be nice if it were more like a static route. Really, I just want to have an automatic way to handle if there is an obstruction or the radio get's moved etc and the signal level tanks. I could just set an access list dropping any low-signal links to keep from having -85 connections on, but that doesn't really help when someone plugs in a netgear monstrosity that nukes the airwaves nearby and I need that node deprioritized in a hurry without dropping them off. I'd rather have one sketchy link for that customer and every other one route around. I can then manually address the noise.
I could just copy the routes from OSPFv3's routes (/routing ospf-v3 route print) to static routes using the OSPF cost as 'distance'. Let OSPFv3 do the work of calculating the routes out but still update OSPFv3 routes based on SNR and CCQ so OSPFv3 does the hard work. Then just copy or update the statics on a timer which don't cause any adjacency drops. Of course, delete non-existent routes in the process. This seems to work really well on my lab in gns3. The OSPFv3 cost changes do their thing up at distance 110 and the statics are the 'real' routes. copy down the changes without any disruption.
The goal here is to get into areas that I can't otherwise. I've tried cambium 450i in 3Ghz and 900Mhz, no go. Can't trench in fiber or DSL. But each home can see a dozen others through gaps in the trees so a mesh is a practical design despite mesh pitfalls. Then, as needed, add 60Ghz links to make a high capacity path with a OSPF cost of 1 for every 60Ghz link. I would also be bridging an EoIP tunnel from the customer's port back to the core router (MTUs adjusted to compensate, 1500 EoIP path).
These are people using hughesnet and that starlink is way to high a price. They're dying for a fair price on a 10Mbps connection so their kids can do homework on.
Finally, if mikrotik would release some terragraph-type gear at a killer price w/ 5Ghz backup then maybe this is all just wasted effort.
Ultimately I'd like to do a dual radio mesh to avoid hop pena