DL vlans and redundancy to Core

[rickpastor]

Hello-

I?ve been given a project to move all vlans from our Core 6509s to the DLs (3750s, etc. MLSs) in various closets, and have read two different ways of doing this:

One way configures two redundant vlans on the Cores, each vlan including one of the two redundant DL to CL links in all the closets, the other Core vlan including the other DL to CL link in all the closets. Besides these two vlans, all the other vlans will be at the DL throughout the enterprise. L2 ports connect the CL and all DL MLSs.

The second way uses p2p L3 links between DL and CL, with separate subnets for every DL to CL link in every closet. EIGRP routes with load balancing.

To this point I have a design following the first method, but would like to know first of all, if this is the recommended method. If it is, then I?d feel privileged to get a couple questions answered about the design. If the second method is generally preferred, then a couple of answers related to this method would be appreciated as well.

On Cisco?s site there are articles supporting both methods, but nothing I could find related to the specifics. Since I?m supporting about 6000 users here, the stakes are high enough where I?d be foolish not to get some advice.

Thank you for any input you can provide.

[tdrais]

I don't think you will get a concencus on this one. The first design used to be the only way to do it because switching was faster than routing. Now with layer 3 switching that is not a issue as much.

It is more which you like better and support best.

I hate fixing spanning tree issues so I prefer to run fully routed. With rapid spanning tree maybe I will have to relook at a layer 2 solution.

[rickpastor]

Sorry for the long delay, and thanks for the response. Can I ask you a couple follow-up questions?

With the first design, is it OK to use access mode ports from DL to CL? If not, would vtp be used?

Also, do I need to configure an SVI on the DL MLSs as well as the CL?

With the second design, is there a concern with the fact of having twice as many subnets configured?

Thanks again for your help.

[tdrais · Accepted solution]

If I read it correctly your first design all the DL switches can see each other directly in 2 vlans.

In this design your core switches only switch they do not participate in the routing. You do not even have to define vlan interfaces for the core switches on these 2 vlans. This is called a switched core and used to be the only way to handle large instalation since it moves a lot of the routing out to the distribution layer. All the DL switches will form neighbor relationships and pass the data in what they think is a direct connection.

Don't know about VTP on a non trunk port will have to try it. VTP is not as important when you do not have a lot of vlans running between switches.

I suspect you will have many more that twice as many. Each point to point will be a subnet. Either way switches nowdays have more than enough memory and you don't care about the size of the router tables sent over the network as much in a lan. If you define your DL switches as stub areas you can reduce the size of their routing table but I suspect it doesn't matter anymore. The main reason to use the second one is all the connection between the switches are point to point so there are no broadcast issues or spanning tree.

[rickpastor]

...yes, my hasty 'twice as many' erroneously referred to the first DL vlan transfer I'll be doing, not the whole picture, as you pointed out.

Thank you for a useful and clearly stated explanation of the two scenarios.