Layer 3 network design

[jdavis1]

Okay, let me bounce this off you guys. I have a campus network consisting of 7 major buildings, and I need to design a backbone for it. All of the buildings are connected by fiber, but it is not a full mesh. If you drew seven squares on a page and connected them all in a circle you would have a pretty good picture of the fiber runs. By patching fiber through some of the buildings I can actually connect my 2 "core" switches which are at opposite ends of the campus to 2 distribution switches in each building.
There are an average of 3 network closets per building and about 3500 ports througout the campus. There are WAN links to corporate in 2 of the seven buildings (same buildings I plan on putting the core switches into), and one of those contains file and print servers for the campus.

C---C
/\ / \
/ X / /\ / / \ / / \ / / \ \
/ / \ D D D D
/ \ / \ / \ / / X \ / X A A A A A A

"C" = core switch - 4500 or 6500
"D" = distribution switch - 3550-12G ?
"A" = access switch - probably 3550
"X" is just a crossover
(Who needs Visio?)

Okay, enough about the physical topology. The current IP design has 4 VLANs based on administrative boundaries. This means that 2 buildings sometimes share a single VLAN. The LAN is layer 2 - all intervlan routing is done by a one-armed 7206.


Here's my question (finally!). I am considering implementing putting 2 3550-12G switches in each plant. I will have each of those connected back to different "core" switches - maybe back to both of them (dual-path). I would like to redesign the IP structure to break it up and to segment it more along physical boundaries rather than political ones. However, if I implement layer 3 in the distribution switches I think I've got to subnet more than I want to, causing administrative hassels, DHCP scopes everywhere, etc. Layer 2 at the distribution layer seems like it would be a lot easier to manage.

What are your thoughts on on layer 2 vs. layer 3 LAN topologies? All of this needs to support VoIP too.

-JD

 

[Piass]

You're definitely on the right lines with that design as far as I can see. You might want to have one VLAN for every couple of hundred users, and therefore a /24 subnet. This might mean that you put the first several 3550 switches into Vlan 10 (for example), then the next few 3550's into Vlan 11. Depending on how the numbers work out, you may be able to have one Vlan per wiring closet.

These Vlans will terminate on the 3550-12G's, i.e. the 3550-12G's will act as the default gateway. You would then configure layer 3 /30 subnets from the 3550-12G's to the core switches. I would also want these links to be more than 1Gbps. If the access switches connect to the distribution using 1Gbps, then you should have something like 4Gbps Etherchannel between the distribution and the core. If you don't do this, you run the risk of creating a bottleneck between the distribution and core. Especially if all your servers are connected into the core,as this is where most of the traffic will be heading. This obviously depends on how much traffic you have going through your network.

You soon then get to the point where the 3550-12G's don't seem to have enough ports and your into using 4500's as the distribution and 6500's as the core.....

Anyway, that's probably how I would do it. It would be a much more scalable design than the layer 2 way. You would end up with more Vlans, but all the scopes could still be provided by one central DHCP server.

If you're looking at 6500's for the core, check out the new Supervisor 720. It's got a 720Gbps backplane...feel the power. Damn, I need to get out more

 

[jdavis1]

Thanks for the input, Piass.

Well, sure, I'd like to put 6500's in as the distribution switches with redundant sup 720s and overbuild the (@#&$@* out of the backbone. I would also love to put 3750 switches in the access layer (if only they supplied power!). However, business is not what it used to be these days (we supply the airospace industry) and there is that nasty cost constraint. Maybe Cisco's Linksys line of products would be more appropriate.

Seriously though, we just don't have the network traffic to justify that kind of expense. I expect that by dramatically increasing the routing capacity by eliminating the 7206 in favor of layer 3 switches will give me a more "bursty" network - more bandwidth used in smaller timeslots. But even if we add a thousand IP phones I don't see us exhausting a couple of gigabits in the backbone. That's an absolutely enormous amount of capacity in this environment.

Speaking of IP phones, doesn't IP at the distribution layer just add complexity in the form of multiple QoS mechanisms? In other words, if I do the routing at the core, then when we roll out VoIP I can simply create another campus-wide voice VLAN and trunk it everywhere. If I do the routing at the distribution layer won't I introduce the problem of configuring IP QoS mechanisms for voice traffic and possibly add cost in the form of a QoS management package like QPM? This is one issue that I am struggling with.

Other than multiple DHCP scopes, what other issues arise when putting layer 3 in the distribution layer?

By the way, has anyone seen Cisco's new Gateway Load Balancing Protocol (GLBP) implementation? I just discovered it. Wish it were available in the equipment I'm looking at. Here's a link if you have a CCO login. There may be a public version somewhere, but I don't know where it is.

http://www.cisco.com/en/US/customer...2/products_feature_guide09186a0080134a35.html