GLBP can be useful but something to also consider with load-balancing between gateways is the layer-2 topology and which link spanning-tree will block. For example if you have a single access-layer switch with dual uplinks to the core and the core has a layer-2 connection between them then one of the uplinks is going to be blocked by spanning-tree anyway so GLBP provides no benefit, in fact for some of the traffic there would be sub-optimal switching because for that traffic it would infact have to traverse the unblocked uplink to one core switch which would then switch the packets to the other core switch which is actually the gateway - in this scenario I use HSRP and LB using VLANs.
Great post and thoughts, I am actually in the middle of building out a Distribution layer that will be running GLBP, or HSRP. My access switches will have L2 uplinks to the dist layer. After reading your post....GLBP is out the window. HSRP it is.
Thanks Chad, appreciate the feedback - glad it has helped you out. It wasn't something that lept out at me immediately when I first read about GLBP - first thought was, wow - thats really cool. Only when started to think a little and thought - hang on!! what about STP??
There are examples in Cisco GLBP documentation on topologies where it might be useful, but to be honest nothing that I would implement on a typical LAN.
Well the solution is to not use a L2 interconnection in the dist layer.
This of course would require L3, which to me is an improvement anyway. As long as you dont have VLANs spanning more than 1 access switch. If you do your stuck with a L2 interconnection at the dist layer.
That's exactly my problem (that and GLBP is not supported on 3560s but that's beside the point).
I have all of my local vlans spanning multiple access switches and thus use HSRP (mostly due to my switch block design). I have been more careful of late (since studying for CCNP) to ensure that the active HSRP router is also the spanning tree root and that the second distribution switch should a) become active for HSRP b) become root for spanning tree, in the event of a failure.
I don't think I could ever use L3 to the access layer, nor a V shaped spanning tree unless my 'access' switch was a badass 6500 or similar. I use a bunch of commodity switches for access layer which translates to a very large modular switch.
Perhaps that is what people are using when the do L3 to the access layer? That or a single subnet per switch?
We currently run L3 at the access in some areas; however as you mentioned, we are using 3750 switch stacks and a single subnet per stack at the access layer.