Cisco Support Community
Showing results for 
Search instead for 
Did you mean: 

Routing implementation at both Core and Distribution layers

If I'm not mistaken, I remember reading some docs saying that routing, in a switched environment, is recommended

to be done at the distribution layer only. The core layer should switch the traffic

as fast as possible, so routing, filters, etc.

are not recommended to be performed at the core. So the question is, is it

really recommended not to perform routing at the core layer?

Are there advantages/disadvantages? Any input is greatly appreciated.

(Based on design docs or on actual experience)



Re: Routing implementation at both Core and Distribution layers

Although the advent of the L3 switches have really improved the performance of the switching hardware, I would still keep my routing decisions at the Distribution layer because the intraVlan traffic at the distribution level will be contained within the distribution layer. Else you would see this traffic all the way upto the core switch which is waste of bandwidth. On the other hand, if you limited L3 switching resources and it is fairly a small network, you could take the routing decision to the core level.

Community Member

Re: Routing implementation at both Core and Distribution layers

I would recommend to switch at the core and route at the distribution layer. As you stated, the purpose of the core is to switch packets as fast as possible. By putting the routing load on the core, it could slow down that layer which could become a potential bottleneck in some cases. Plus, with local subnets tied to the distribution layer(say we had two subnets off of one switch), it would be a mistake to do things like send traffic from one subnet on the distribution layer switch up to a router and then back down to the same distribution layer switch to be sent to the destination subnet. If you need to route at the core and are using a multilayer switch, there are some great techniques that improve switch performance. One idea is netflow switching which is implemented in Layer 3 switches.

With netflow switching, the first time traffic needs to get to its' destination, it will go up to the supervisor engine in the switch so that it can be directed toward its' destination. From then on, this added step is bypassed and traffic can go directly toward its' destination. Also, if you are going to use a router at the core, if it is a chassis based router like a 7000 or 7513, if you can afford them there are processors that can offload some of the work of the router such as VIPs.

I agree with a previous reply to your question that if it is a small network, you could get away with a router at the core. Yet, I do not know what kind of network you are dealing with, so I am just giving some general information to you that I hope helps!


Re: Routing implementation at both Core and Distribution layers

First of all, thanks for the informative input. Basically I'm looking

into a large campus network design. The link below shows a L3 backbone

design. It was mentioned that this design is usually used when very high

performance for handling multimedia applications based on IP unicast and

multicast is desired. And comparing this with a L2 backbone, I've noticed

the core-to-core link was not used. Does that mean for L2 backbone design,

core switches should not be linked together, always? What could be main

impact if the core switches in a L2 backbone design were linked together?

And for L3 backbone, what could be the main advantage of linking core

switches together? Sorry for the number of questions but any info you have

is really appreciated. Thanks again.

CreatePlease to create content