3850's as routers

[sw43]

revious we were taught to switch when we can and route when we have to.

Now a days I understand with fast L3 switches we can route on the fly.

So in my small LAN of (9) 3850's with a 3850-24XS as the core router I want to abandon L2 VLANS and put my 3850's into L3 mode with no switchport command.

I envision behind each 3850 servicing a floor of ~20 workstations they are there own IP subnet so I'll end up with (9) IP subnets.

At the core "switch", will config each interface w/ no switchport and IP'd as the d-gateway for each "upstream" access switch.

Which leads to my question do I need a routing protocol defined or since the "access" 3850's are "C & L" (output of show ip route) directly connected to the core they route automagically?

[Reza Sharifi]

Be careful with this type of design as you will not be able to span vlans across multiple switches. So, for example, if you have 5 printers and each one connects to a different access switch, each printer will end up in a different subnet. So, before you make this change think about the areas that you may need to extend vlans from one switch to another.  in my opinion, you don't need to change the current design but if you really want to do it, you can run a routing protocol on the access switches of simply a default route from each access switch to core.

HTH

[luis_cordova]

Hi @sw43,

With the command:  S(config)# ip routing you can enable routing in the switch L3.

Then check the result of the command: S#show ip route

If you can see the configured networks, then you will not need a routing protocol to communicate with each other.

If you need those networks to be seen from other networks, you will need a routing protocol.

Regards

[Richard Burts]

The original poster has not provided much detail about this network and that impacts our ability to understand the situation and to give good advice. We are told that there are 9 access switches and a core 3850 and not told anything else about the topology. I am assuming that the 9 access switches connect to the core 3850 and do not have any other connection outside of their building. We are not told whether each building is already its own vlan or whether the network is currently one flat network. If there are existing vlans we are not told whether any existing vlan spans between buildings or whether every vlan is contained within a building with one connection to the core. 

If the network is one flat network then there probably is an advantage to creating separate vlans for each building which reduces the size of the broadcast domain.

If every building already has its own separate vlan then I do not see that it makes much difference whether the connection between the access switch and the core switch is routed or is switched. Any device communicating with another device in that same building will be locally switched by its access switch (no need to go through core). And any device communicating with a device in a different building will need to be forwarded to the core and then be forwarded to its destination. That continues to be true whether the forwarding decision is a layer 3 decision or a layer 2 decision.

If the decision is made to make the links between access switch and core switch into routed links it will reduce the size of the mac address table in the core switch. In some networks that could potentially be significant. Given the described size of this network I am not convinced that the savings on the core switch would be significant.

HTH

Rick