L3 switches-> Efficient bandwidth utilization, for Campus network

this should be in the same set of white papers - try the reference designs at

it probably only matters if there is resilience, where the issue is about making use of redundant links when you dont have a fault.

A layer 2 campus with resilience needs to use spanning tree to prevent active loops - and any blocked port isnt carry traffic.

frankly i think this is probably a red herring, since bandwidth efficiency in a campus isnt that important - more ports in the core, high speed links and extra fibre are all typically fairly cheap anyway compared to overall build cost.

much bigger issues are reliability and convergence time. once you are carrying traffic that doesnt tolerate long interruptions in forwarding (IP Telephony for example) then L3 is difficult to do without.

Well, I'm not exactly sure the reference that you are speaking of, but here are my thoughts...

Many campus networks now have a need for a lot of different vlans and there is a need to route between the vlans. You can develop an L2 scheme with everything going back to the core to be routed. However, it would be much more efficient to have distributed routing at the campus. Given that campus networks now have high bandwidth usage, this could prove to be an extremely expensive using routers. However, L3 switches can be much less expensive, have many more ports and can forward the packets at a higher rate. So, it only makes sense to deploy a handful of L3 switches instead of routing everything back to the core or using multiple routers.

Hope that helps,

Jim

I agree with Jim. Distributed layer 3 can be very rewarding, and is also the latest push from Cisco for campus datacenter networking. That being said, things get complicated in that distributed layer 3 networks require some GREAT IP/Network planning, as you really need to focus on summarization. In the core model, you can create large vlans and trunk them out and slowly fill them up as servers are added. In the case of distributed layer 3 (or at least true, fully distributed layer 3), each subnet can only have as many IPs as there are ports. Not only that, but I'm sure you are going to end up with requirements for servers, or load balancing, etc, that will require segmentation. So how do you decide what your model for VLANs are going to be on the l3 switches?

Perhaps a good medium solution is to roll out Layer 3 to each building, but still have a set of 'core' switches in each location that hold your layer 3. This way each building will have layer 3 and routing relationships, but you don't have to worry about potential segmentation nightmares. In short, there are 100 ways this stuff can be done, but it all comes down to your current requirements, your knowledge/understanding of changing/future requirements, and your budget. Pushing cores to each building may be costly, but then again you can use cores as distribution and access layer switches if you don't have ungodly requirements for bandwidth.

This is why it is hard to find any kind of bible for 'network design', because there is no single right answer, and just about every business has vastly different requirements.....there are always exceptions.

That's why we get paid the big bucks, right? :-P