Spanning Tree Protocol question

[BespinReactorShaft]

Is there any critical problem when STP (in particular the Rapid version, RSTP) is used in a Layer 2 (Eth) ring network? i.e. as opposed to a meshed topology?

There are talks of a Ring STP (STP optimised for ring topologies) but I understand that this will probably never take off due to the introduction of RPR 802.17.

But if we really must insist on a non-RPR Layer 2 ring network, can we still go ahead with RSTP?

Thanks.

 

[spidey07]

well, in this day and age we try to avoid spanning-tree loops if at all possible.

With the advent of cheap layer3 "switching" there isn't much need for spanning-tree at all.

So the next question is "do you really need spanning-tree?"

Because the control and quick convergence of routing protocols dominate any real need for it. Sorry if I didn't address you particular question, but quite frankly spanning-tree is ancient and a pain in the rear.

 

[BespinReactorShaft]

I thought STP is supposed to prevent loops in Layer 2 networks.

And, no, this network will not have any Layer 3 functionality. Pure Ethernet MAN between several geographically distributed LAN sites. We are looking at a means of protecting L2 traffic within the confines of this Metro Ethernet ring.

 

[Goosemaster]

Originally posted by: ming2020
I thought STP is supposed to prevent loops in Layer 2 networks.

And, no, this network will not have any Layer 3 functionality. Pure Ethernet MAN between several geographically distributed LAN sites. We are looking at a means of protecting L2 traffic within the confines of this Metro Ethernet ring.

EDIT: I didn't read it correctly.....are there ANY routers at all though? Even RIPv1/2 being used at the ends would do the trick.

 

[Goosemaster]

Where is the traffic going? Is it a private line or is it used for external/internet access as well?

If so, using a router and a layer three switch at some point would allow for a simple routing protocol if you do not want mess with clients' data. In addittion, the router would allow you to use VLANs to each endpoint which would take care of security.


I am a novice at this so please keep that in consideration. STP would definitely prevent the loops, but I would think that VLANs would be useful for security reasons. Like I said, RIPv2 will do it, but obviously for such a large application, I am sure you would use something better such as OSPF depending on the equipment at each juncture.

In addittion, you could use VTP if you needed it if they have to communicate with each other via the MAN.

 

[Goosemaster]

Also check out RSTP and RRSTP 802.11w's.


My question to Spidey: Is convergence time really a problem on a MAN though? I assume most every client uses a router. Unless they love popping it off and on or power service in the area isn't why would the network have to converge again? Even if it is due to foolish construction workers, its not like it would be a daily occurence on such a tightly knit network.

 

[spidey07]

Originally posted by: Goosemaster
Also check out RSTP and RRSTP 802.11w's.


My question to Spidey: Is convergence time really a problem on a MAN though? I assume most every client uses a router. Unless they love popping it off and on or power service in the area isn't why would the network have to converge again? Even if it is due to foolish construction workers, its not like it would be a daily occurence on such a tightly knit network.

Its actualy layer2 loops that are the problem to begin with (most design rules these days say avoid them at all cost). There isn't much control with spanning-tree. Sure there have been advancements as mentioned but there is still the basic problem of "no control"

It will work, just make sure you know where your roots are and actually plan the tree accordingly. But I'd highly recommend using a layer3 model (there is hardly a price difference between layer2 and layer3 switches these days - all sorts of goodies like rapid convergence, redundancy and load balancing become available.

 

[Goosemaster]

What about MPLS? That serves to "route" ethernet traffic, but does it serve to prevent loops in a ring?


MPLS is not foreign to me, but rings are like Latin to me 😀


<---does not speak latin🙁

 

[spidey07]

Originally posted by: Goosemaster
What about MPLS? That serves to "route" ethernet traffic, but does it serve to prevent loops in a ring?


MPLS is not foreign to me, but rings are like Latin to me 😀


<---does not speak latin🙁

MPLS is a layer3 technology with pretty darn good traffic engineering as well. It relies on the underlying routing protocol to make path decisions.

😉

A ring is just a ring topology where you have a bunch of switches well...in a circle/ring. Common in MAN scenarios where fiber routes aren't cheap. Then again there's always SONET with is what most MANs are built with. Has its own self healing ring topology.

 

[cmetz]

802.1w rapid spanning tree is only really rapid in the case where you have a point to point link, both sides are properly configured 802.1w implementations, and both sides are up. In most real world scenarios, you will only get faster convergence some of the time, and the rest of the time you're stuck with 802.1d style convergence.

Spanning-tree is based on this fiction that your bridges (switches) are in a strict heirarchy going outward from a root bridge. That doesn't map too well to a ring. In fact, I think a ring might be the pessimal scenario.

RSTP won't really hurt you, though. I just don't think it will help as much as you might be led to believe.

Spanning tree, in my opinion, is NOT a good choice for any type of routing or failover protocol. My experience is that it doesn't converge fast enough to give you happy users, and it often behaves in ways that are not obvious to someone who hasn't spent a lot of time working with spanning tree (read: the folks you might have doing day to day support). In my opinion, spanning tree's use in a modern network should be confined to a bit of protection against shooting yourself in the foot - accidental or unintended loop creation. STP is still good at that.

OSPF is easy, and handles redundant links and failover well. If you find a way to reorganize your network into a routed network, you'll be better off in the long run doing things that way.

 

[BespinReactorShaft]

I'm concluding here that STP or any of its flavours will of course work, just not in a very optimised manner to provide failover switching in a pure L2 ring network. Now if only we could convince this operator to buy our RPR story....