Netonix Forums

I have a complex WISP. Okay, perhaps it shouldn't have been complex, but thats where we are. And perhaps it should have been routed, but I've found Netonix switches to be a damn sight more stable than Mikrotik Routers.

So - 30+ transmission sites, some of them larger than others.

I'd like to plug in any old device, joining each site. And I'd like to plug in a secondary any-old-device too.

So switch at Transmission site 1 - for instance, has two links - port 2, and port 3 - with Transmission site 2.

First thing I'd like is that the untagged VLAN is only shoved over one, till ping test fails, then it's switched over the second. So a rudimentary failover. On a vlan by vlan basis, so I can still manage the secondary link devices.

Second thing is to somehow bond the two links into one, so that we get both bandwidths (and thats far more difficult, etc).

Would this even be feasible?

Or do I have to hold my nose, put on my big-boy pants (and boots) and wade through the Mikrotik Swamp again?

---* Bill
http://www.marykirk.com

The VLAN thing I would not hold my breath for.

THe second thing about bonding wireless links we are kicking around but not happening in the "near" future

Do your self a favor....

Buy a truck load of EdgeRouter X routers. Play around with 3-5 to learn OSPF. Then when you're ready, OSPF + BFD will be ready in firmware v1.8 (v1.7 is actual firmware version). BFD is easy to learn, just a 1 command and you're of...

Start deploying the routers (You can power EdgeRouter X with the switch).

Then you'll have a very nice routed network with automatic failover....

Need capacity ? Forget about bonding. Go buy some AirFibers instead....

Just my 5 cent :-)

^
routed is MUCH simpler to deal with. It's a more complex layout, having to remember different networks or do DNS etc, but troubleshooting is MUCH easier.

OSPF can be used to handle redundant links just by setting weight. Also, you can do ECMP so if two links have identical metrics and OSPF has them both as active, you will get some quasi bonding via flow. behavior is somewhat like LACP but layer3. Not all platforms handle ECMP, but the linux kernel does and EdgeOS is able to as a result.

Damn and I have five Edgerouters I was going to stick on eBay... Oh well. ;)

Thanks!

---* Bill

so the upcoming 1.8 EdgeOS is a 'lot' to look forward to. I can share what's already public, which is a completely new (well, new to edgeos) routing stack with MPLS, VPLS, BFD, and a new queue tree, and some new bonding tricks *AND* and wizard for bonded wireless links ;).

Edgerouter Lites should be able to handle a pair of AF in a layer3 bond pretty well.

I have a lab of ERL running on a 1.8alpha build and I'm quite happy with the changes so far. Very exciting to have a replacement for mikrotik that doesn't break the bank.

as far as your VLAN scenario. I don't see a good solution to this as-is. This is where MPLS+VPLS(optionally MPLS-TE) really shines. You can tunnel a VLAN interface on the tower side to a VLAN or port anywhere else via a VPLS tunnel and let the label paths handle your redundancy -or- specify the path and the alternate path ahead of time. pseudowires if you are familiar with cisco. you can do MPLS-TE+VPLS between routers, you don't have to do it network wide.

for load balancing
LACP works with VLANs, it's not perfect balancing between two links. mac pairs stick to a leg in the LAGG.

MPLS-TE also has some tricks. You can have a primary and secondary path. You use up the primary first, then it starts using the secondary as an overflow. It's not a balanced sharing, but does allow you to take advantage of the throughput. You can 'balance' this by pushing certain types of traffic down each link but letting them overflow (via setting secondary path) to the opposite. what I mean by not balanced, is that you set a primary path and a secondary. you will use the primary only until you hit the pre-determined capacity, then use the secondary. If you have 2 100Mbps links and are pushing 150Mbps. The first link would have 100Mbps and the second would have 50Mbps.

ECMP (equal cost multi-path) handles IP transit by rotating flows in a round-robin like dispersion. this isn't round-robin per packet, it's more like LACP where a single flow sticks to one link for it's lifetime and the next new flow goes to the other link.