On 2/25/20 10:48 PM, Abhi Devireddy wrote:
L2 rings IMHO seem pretty brittle. I know there are L2 ring products like Juniper BTI, which use ERPS and not strictly STP/RSTP to move blocking ports, and those seem a little better although it's mostly statically configured.
For a strict ring topology like this, I'd certainly consider E-RPS or similar over [R]STP if you were going to do this all at L2. It's a lot more "predictable" in my experience insofar as it's harder to shoot yourself in the foot by mistuning some knob somewhere and getting behavior you don't expect.
That said, I'd be loathe to put 30+ switches on a ring even within the same building unless I had little other choice.
One thing that I haven't seen explored in this thread is the idea of doing things at L3. If you've got 4 fibers, you can use Bi-Di optics to build a partial mesh/ladder/braid as you go up the building. That can be a mess at L2 with (R)STP, but carving out an IGP area and doing it at L3 is often not nearly so ugly. If you've got L3 switches (which are cheap, these days), it may be a good option, though it may also be annoying from an IP subnetting POV unless you overlay it with something like an IPv4-in-IPv6 core (MAP, 464XLAT, etc.) or an L2-in-IP overlay like VXLAN both of which substantially increase the conplexity of the situation.
Using CWDM or DWDM with 8-16 channels on either 2 or 4 trunks across your 4 fibers to do a more conventional home-run with multi-chassis LAG or similar is another reasonable option.
I'd avoid stacking 30+ switches even where you have stacking support over fiber especially if the switches aren't in a physical stack. Too much opportunity for split-brain scenarios IMO.
I'd contribute to the "see really hard if you can just drop more fiber down the riser" echo chamber.
-- Brandon Martin
