Ring organization. How right?

Question

Cisco

Good afternoon. There is a geographically distributed network (within one enterprise). In the above diagram, all switches are interconnected by optics (each switch has 2 SFP ports 1Gb/s).

Task: to organize a fault-tolerant ring, i.e. in the event of a break in any connection between the two switches, the network should continue to function properly. VLANs are used.
At first glance, you can use MSTP, and everything will work. On the test bench, we got a convergence time within a second or two. However, studying the discussions on STP (on Habré), I came to the conclusion that this approach is not a cake. Actually the question is how to do it right?

Answer 1

На вашей схеме MSTP излишен, достаточно RSTP. Кроме того, выставленные Hello в 1 сек и настройка протоколов семейства BFD сделает сходимость быстрее (максимум пропадет 1 пинг). Остальное для вашей картинке от лукавого.
Если вас интересуют еще технологии, позволяющие резервить кольцо на L2: mlag (hp), vpc (cisco), errps(dlink), vrrp(huawei). Как видите, все они вендорозависимы. А вообще старайтесь уходить от резерва на L2.

Answer 2

MSTP and do not soar your brain with holivars from habr, have you personally encountered problems? If not, why do you think that your crutch will be more reliable.

Answer 3

those. in the event of a break in any connection between the two switches, the network should continue to fully function

partial-mesh/hybrid-vpn?

Answer 4

rpvst+ (it seems that this sounds like a tsiska), but in general, what is wrong? The scheme is a classic ring, there is no need for more

Answer 5

In the above diagram, all switches are interconnected by optics.

In addition to 3750, which switches (vendor, model) are present in the ring, please explain. Is the network monovendor?

At first glance, you can use MSTP, and everything will work.

Can. Or you can use Rapid-PVST+. In the second case, everything will be simpler and clearer, in my opinion. But Rapid-PVST+ is a vendor specific technology. But on the other hand, if you want to set up STP (MSTP, for example) in a multi-vendor environment, then be prepared for the fact that you will have to diagnose problems that should not be "according to the book", but they are.

On the test bench, we got a convergence time within a second or two.

How much do you need? If you choose between downtime per second and several hours (if the link is broken and there is no reservation), then the choice is obvious, in my opinion.

However, studying the discussions on STP (on Habré), I came to the conclusion that this approach is not a cake.

STP is the kind of thing that everyone with CCNA knows about, but it has a huge amount of nuance and not many people understand it already. This should be taken into account when assessing the usefulness of the opinion of a stranger to you. Also, not only is your approach "not a cake", but STP itself is "not a cake". If you look at things in development, then the very idea of ​​\u200b\u200bethernet bridging is not a cake . But there are some ideals, and there is a "reality" with which to work.

Actually the question is how to do it right?

I don't know and no one knows. In general, the very idea of ​​\u200b\u200b"right" / "wrong" decisions is not entirely clear to me. There are working solutions, there are non-working ones. There are solutions with minor side effects, on the contrary, with tangible ones.
If I were you, I would set up Rapid-PVST +, tweak the timers if necessary, set the priority of the root switch hard (if I understand correctly, 3750 is suitable for this role in the topology) and work out typical accidents (all this in the lab), I would not forget about udld. On the other hand, if in the foreseeable future it is possible to add equipment from other manufacturers to the ring, or the number of vlans is too large, it is worth considering MSTP.