Layer 2 links to sites need redundant loop

[DocSkyhawk]

We need to build a layer two network between buildings and so far I have done some mock ups in PT and have some issues with the way we would like to design it.  

Requirements:

Layer 2 all on the same subnet 

DHCP requests need to work at any location back to the core location

Two links to each location but not two directly to core(think serial wiring not parallel)  

What we have tried so far:

Same port channel on each link in a "circle/loop".  This causes a storm and the traffic ends up not working(On mode)

Different port channel on each link - same as above(not really a surprise)

Port channel in the active mode in same loop, one link goes down much like STP and when a packet leaves the source, it often goes the "wrong" direction in the loop and dies at a dead end.

Most likely working setup - normal trunk links and let STP kill one link and everything works great.  The issues here is that I haven't been able to figure out how to pick which link goes down and all the bandwidth in the loop comes from one side potentially overloading the links near the core.  If we use two cores, I would prefer to kill the links to the two cores so traffic can travel down both sides of the loop at the same time.

Thoughts?

[luis_cordova]

Hi @DocSkyhawk ,

In my opinion, keeping the layer 2 network in a ring shape contracts the problems you describe.
My recommendation would be to configure some of the MultiLayer switches with SVI and some of their interfaces as routed ports.

These measures will prevent layer 2 loops.

Regards

[Seb Rupik]

HI there,

If you just want to block the link between the two cores, then configure a STP cost which is higher than the sum of all the other costs in the ring.

Say you are using VLAN ID 2, and core 1 and core 2 are connected via Gi0/1 :

!! Core1
!
spanning-tree vlan 2 root primary
!
int gi0/1
  desc link to Core2
  spanning-tree vlan 2 cost 1000
!

!! Core2
!
spanning-tree vlan 2 root secondary
!
int gi0/1
  desc link to Core1
  spanning-tree vlan 2 cost 1000
!

This should insure the inter-core link cost forces the STP calculation to put it in a Blocking state.

...but if you are running a FHRP like HSRP this will lead to a very inefficient forwarding paths for location 4!

cheers,

Seb.