3

topology

Hi, what would be the best practice to achieve separate tree topologies for each of the rings? Two options come to my mind, although I do not completely understand IF RSTP can be made VLAN-aware somehow (I was always under impression that it is VLAN-agnostic).

Each of the transport VLANs in each of the rings is labelled as "2" with ports on edge devices labelled as "22" to achieve separation between VLAN 2s on L2 level. All ports untagged.

  1. Three MSTP instances (one for each of the rings: A, B and C) - VLANs 2/22 mapped to each instance.

First question here - should MSTP instances map different VLAN numbers to them (e.g. relabel VLAN2 in ring A to VLAN22 for MSTP instance 1 and VLAN 2 in ring C to VLAN222 for MSTP instance 3)?

  1. VLAN2 relabelled as VLAN22 in ring A, VLAN2 relabelled as VLAN222 in ring C - all three VLANs tagged in each respective physical ring, enabled RSTP.

second question here: RSTP is supposedly VLAN-agnostic - it simply operates on ports and BPDU broadcasts - will such an idea as described above even work?

1 Answer 1

5

what would be the best practice to achieve separate tree topologies for each of the rings?

Forget the rings and reconnect the switches in a tree topology with one or two center switches.

This double-rooted tree topology has no single point of failure. Any access switch failure breaks connectivity only for hosts connected to that single switch. Any host connecting to two access switches has fully redundant connectivity. enter image description here

A tree is a more natural way to arrange an Ethernet network. It keeps paths shorter, removes bottlenecks, has a smaller diameter and provides better resilience. Of course, you can forgo redundancy and use a single core and single uplinks. Even a simple tree has just the core switch as single point of failure. Your ring system has two SPoFs (the outer switches in the light red box).

In any case, make sure your root bridge is always well-defined (one of the center switches).

I do not completely understand IF RSTP can be made VLAN-aware

RSTP is entirely VLAN-agnostic and works on the port level only.

You can use MSTP with multiple instances and grouped VLANs to overcome that limitation. If you must stay with the rings you could use instances to optimize link usage, but note that an instance border is never in the middle of a link (as pictured) but inside a switch (between VLANs).

Note that the topology in your diagram exceeds the design limits of RSTP and single-region MSTP (maximum diameter of seven bridges) and will become instable if the 'wrong' node fails (the diameter prevents proper convergence). You could overcome that limit by running multiple MSTP regions.

All in all, everything is going to be much simpler in a proper tree topology. Also note that the leftmost and the rightmost RING-B switches are single points of failure, cutting off the lateral rings.

The double-rooted tree pictured above has very short paths and no need for configuring multiple MSTP instances or regions. Optionally, you could optimize link usage by configuring MSTP instances, then it's also better performing than the chain-of-rings topology in your question. Of course, IEEE 802.1aq Shortest Path Bridging is almost always superior to xSTP.

13
  • Thanks, you've been very helpful. What about MSTP instances then? Should they all contain separate VLANs (i.e. instance "1" for ring A would contain VLAN22, instance "2" for ring B would contain VLAN2, while instance for C would contain VLAN222), or can they be left as they are now (VLAN2 everywhere, with exception of VLAN22 at edge ports, one spearate instance with identical VLAN numbers for each of the physical rings)?
    – WoyoC
    Feb 15, 2022 at 11:29
  • Exceeding the design limits, instances don't really help since all instances within a region span the latter. Instances are only used to create a spanning tree each, but propagation of BPDUs only stop on a region border.
    – Zac67
    Feb 15, 2022 at 11:50
  • Good answer, but I'd offer a disadvantage of a tree over a ring. Since there is no core switch a ring never "runs out of ports on the core". Contrast this with a tree, where you can only add so many nodes before the core switch runs out of switchports. Nov 15, 2023 at 18:30
  • 1
    @Zac67 that was the point of my comment... and I would point out that in my experience core scalability is the biggest problem people face in a design. In the beginning, the architect has to be smart enough to know how big to make the core router / switch. If you choose a ring, you can effectively circumvent the core scalability problem (although a tree provides much more bandwidth per access switch) Nov 17, 2023 at 12:18
  • 1
    IMHO, this winds up being another "it depends" situation. Most of my experience backs up what you say; if all your uplinks are contained within a building, adding new uplinks isn't huge. If you need to call a construction crew to lay out new uplinks, a ring suddenly has much more appeal. Nov 17, 2023 at 13:11

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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