3

We have 2 buildings which are interconnected by 2 wireless P2P links. One is used as main connection, the second one is for redundancy and backup. So far, we have these 2 buildings in flat L2 network and each building A and B has 1 "core" switch where wireless APs are connected. These 2 APs are different vendor and different speed ( Mikrotik and Alcoma ). Simple diagram should explain it a bit better: diagram

We bundled those 2 links in PAgP because those wireless links are transparent for the switches and link between Catalyst 3560 and Catalyst 2960S is shown as 1000Mbit full duplex. Mikrotik is using 802.11ac, finding more info for Alcoma is somehow difficult, because it's old model and I couldn't find any info from manufacturer. All I know is that it works in 24GHz band.

The question here is, should this even work? From what I know, PAgP and Etherchannel works over Ethernet links and those links must have the same speed and duplex settings. What are the outcomes of this little "cheat" using transparent bridges? We noticed, that from time to time bundle is broken on Mikrotik port ( Gi0/39 ) but after issuing shutdown/no shutdown, bundle is successfully created again. I am suspecting that some timers in PAgP are not met or the link is not as stable?

Could this work without PAgP and manually setting up Etherchannel?

Thanks for insight!

  • Thanks for the information @RonMaupin. We were thinking about throwing away negotiation protocol, we will definitely try it out. As for your side note, I agree. Currently we are working on redesigning infrastructure by using more VLANs and ending those VLANs at Catalyst 3560 as it our "collapsed core". Problem might be, that we don't have another true L3 switch in Building B. – Lukas Sep 9 '15 at 5:08
  • If you use LACP instead of PAgP, could you not use port priority and get most of your data sent through one port over another? – Jordan Jul 1 '18 at 0:10
3

Yes, this should work since the channel is switch-to-switch ethernet from the perspective of the switches.

If you don't actually need to negotiate the channel, why use a channel negotiation protocol? That's just one more unnecessary complication. With Cisco switches on each side of the channel, why not just use the channel-group <num> mode on command. You don't seem to need channel negotiation.

On the positive side, failover with channels is faster than even rapid spanning-tree.

On the negative side, some of your traffic is traveling on the backup link, which may be sub-optimal.

As a side note: I'm not really a fan of trying to have a layer-2 domain across something like this. There are very few things that require sharing a layer-2 domain; most things today can be layer-3 connected.

3

The big problem with this is out-of-order packets and statistically half your traffic flowing across a slower, more loss-prone link.

If you can do L3 (ip) across both links, I would suggest "CEF per-packet" load balancing. But again, half the traffic would be crossing a slower link. (I'm not sure CEF uses link bandwidth.) What you have is an unequal cost multipath network.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.