[c-nsp] MST Experiences: was Re: Dell switches (specifically PowerConnect 7048P) and Ciscos

[Phil Mayers]

On 11/28/2012 09:13 AM, Saku Ytti wrote:
> On (2012-11-28 09:55 +0100), Gert Doering wrote:
>
>>> I'm sure there are topologies in which mst is suitable
>>
>> Textbook topologies, obviously :-)  - where you sit down, design your
>> network, implement it, *and then go elsewhere* instead of modifying
>> your network on-the-fly.
>
> We've ran MST as long as it's been in IOS. Mostly in metrorings. We have 4
> instances, for not particular reason but somewhat even VLAN divisors. But
> they all have always been using same topology. And we never touch the MST
> config, nor have ever had the need to do it.
> We could do some micro-optimization to make sure that metroVLAN which is
> terminated to eastPE has eastPE as root and meteroVLAN which is terminated
> to westPE has westPE as root. And we could then define 1-2000 are westPE
> terminated VLANs and 2001-4000 are eastPE terminated VLANs. But we're not
> even doing this, if we'd want to do it, we wouldn't need to touch MST
> config.
>
> I know I don't see eye-to-eye to this with Gert, but I think MST is
> perfectly usable in most networks if you design the network and you have
> system to allocate VLANs.
>
> Quite few networks actually need unique topology per VLAN without it being
> hard-to-observe micro-optimization. But there certainly are some how can
> capitalize on it.

I certainly don't *need* per-vlan topology. But I can use it at no 
observable cost.

On the other hand, I could have re-numbered every vlan on every switch 
at all of our sites, spent days or weeks planning vlan numbering to 
ensure there were enough tags allocated to each space/zone/thing, mapped 
the 15 instances IOS uses to the >45 definable topological areas, and 
then watched it all break weeks later as the first "special job" came 
along and I had to incur a topology change network-wide.

So, for us, MST is pointless.