Core Design Help Please!!

Unanswered Question
Jun 12th, 2007
User Badges:


We currently have 2 x 3550 t Switches as our core and layer 3 switches.

The stacks are a mixture of 3560's and 2950's linked via sfp + fibre cables. The main core switch is the main for hsrp and the 2nd core switch is the backup. We have costed the backup ports on each stack switch and also the redundant switch so that the main route is always preferred.

My boss has said that he thinks we should have our larger stacks of 3560/2950 switches split their traffic across our 2 core switches by changing the spanning tree costs for these backup ports making them more attractive. He wants this as he says it does not utilise our other switch. Which is true.

ie have the top 2 switches send packets up to the main core switch and the bottom switches send packets to the backup switch. This would mean re-costing all the spanning tree as the backup hsrp switch has costings on each port of 5000 so that it is never the preferred option unless the primary fails.

Maybe we should be using a load balancing method instead of HSRP?

Now I don't really see any benefit in doing this only to create more work and confusion within the network.

We also have a flat network and I have started to create new vlan's for dmz's and remote sites but I would think that we should be creating vlan's for our servers etc, the servers plug into various ports on one of the stacks and I would think that our app servers should be plugged into the core but that is full with our stack connections?

Please take a look at our diagram and let me know your thoughts on our design and improvements we could make.



  • 1
  • 2
  • 3
  • 4
  • 5
Overall Rating: 0 (0 ratings)
Amit Singh Tue, 06/12/2007 - 07:07
User Badges:
  • Cisco Employee,


If you are using one switch to route all the traffic then you are putting an extra load on the primary HSRP switch.

You can achieve this by using MHSRP (Multi HSRP) instances on both the switches. This can only be achieved if you have multiple vlans on your network. if you have a flat network you cannot achieve the HSRP+STP load-balancing.

Just to give you an example for your core switches 3550.. if you have 10 vlans on your network then you can configure the SWITCH 1 to be the spanning-tree primary root for vlans 1-5 and spanning-tree secondry root for vlans 6-10. On the SWITCH2, configure the vlans 1-5 as secondry root for spanning-tree and vlans 6-10 as the primary root for spanning-tree.This will take care of the layer-2 STP load-balancing.

For layer 3 load balcning configre the SWITCH1 as HSRP master for the vlans 1-5 (the same vlans which are primary for STP) and standby for vlans 6-10. On the SWITCH2 configure the vlans 1-5 as standby and vlans 6-10 as master (the same as on layer2 stp). this will take care of the layer3 load-balancing. This way you both the layer-2 links from a particulaer stack will used for load-balancing and forwarding the traffic.

Please see the linke below for STP config:

Link for HSRP:

HTH,Please rate if it does.

-amit singh

restoreml Tue, 06/12/2007 - 07:35
User Badges:

Thanks for your advice.

Yes we are using only the primary HSRP switch for all traffic. We will start to implement more vlan's soon so I will us hsrp + stp.

Would anyone recommend what my boss is asking for and that is for the top switches in the stack to go up the stack for their traffic and the bottom switches in the stack to go to the standby core switch if we only have 1 vlan for all our users at the minute? I am not too sure that stp will run this efficiently anyway as he suggested?


chris.lepa Tue, 06/12/2007 - 10:23
User Badges:

I know that GLBP can load balance gateways on a round robin basis for the same subnet... Thing is, I'm not sure if your layer 3 switches (or any at all for that matter) support this protocol though. You should look in to see if your switches support GLBP, if so, that would be a good temporary solution.

chris.lepa Tue, 06/12/2007 - 07:11
User Badges:


Does your network include multiple VLANs? If so, then you CAN load balance using HSRP. Just create SVIs on each core switch for each VLAN, then put some in a standy state and some of them in an active state. Then on the other switch, make sure the opposite is done.

EX: SW1: vlan 1-3 active

vlan 4-6 standby

SW2: vlan 1-3 standby

vlan 4-6 active

I hope this gives you some idea of what to do.


nate-miller Tue, 06/12/2007 - 10:37
User Badges:
  • Bronze, 100 points or more

But unless spanning tree is configured properly, in order to get from a leaf up to the core 2, it might have to send packets up the uplink to core 1, then across a inter-switch link from core1 to core2 (if it exists.)

If there is a trunk between core 1 and core 2, then you'll need to play with spanning tree and figure out where all our spanning tree blocking is- if the blocking isn't at the trunk between the core switches (if one exists!) , you're going to have non-optimal behavior.

The easiest way to do this is an even/odd distribution, but you can do vlans by range, ro whatever.

Start with spanning tree:

On core 1:

spanning-tree vlan 1,3,5,7,9 root primary

spanning-tree vlan 2,4,6,8,10 root secondary


interface range vlan1, vlan 3 , vlan 5, vlan 7 , vlan 9

standby priority 110

standby preempt



And on core two you flip the spanning tree, and raise the priority on the other HSRP interfaces:

spanning-tree vlan 1,3,5,7,9 root primary

spanning-tree vlan 2,4,6,8,10 root secondary


interface range vlan2, vlan 4, vlan 6, vlan 8 , vlan 10

standby priority 110

standby preempt



GLBP would work too, but you'll still need to be careful of your design with respect to spanning tree.

(My examples are for native IOS, 12.2SXF. You may need to play around for your version of code, but the idea is there.)

chris.lepa Tue, 06/12/2007 - 11:57
User Badges:

Hi Nate. HSRP would definately be the best solution, but he only has one VLAN, so it couldn't be used for load balancing purposes. This is why I suggest something like GLBP.


restoreml Wed, 06/13/2007 - 00:27
User Badges:

So if we can start to separate into vlan's then hsrp is going to work fine. If not then we can use glb.

So no-one would suggest that we split a stack using spanning tree and have both core 1 and core 2 used equally that way? Would anyone ever do this? My boss thinks that it's too many hops for a stack of 5 switches to all route back up to the top switch for routing purposes instead of using the 2nd switch as it's closer then the gig trunk link between both cores. I don't think that the gig fibre connections all the way up the stack from the bottom switch would have any latency.

I am not sure that the 2950's will support this but i'll have a look through.

Thanks to all


chris.lepa Wed, 06/13/2007 - 06:07
User Badges:


Your boss's STP plan sounds too unnecessarily complicated. You can't load balance layer 3 routing with STP, you can only load balance root bridges for VLANs.

I recommend that you split this architechture to include two vlans (or more, if you feel like really getting organized) and use HSRP. And if that is out of the question, then look into GLBP.

Oh, and by the way, your boss is wrong about there being too many hops. A hop only occurs when a router is traversed. There are no hops as traffic traverses a switch stack.



This Discussion