VPC vs stacking Layer 3 fowarding

[jmprats]

Hello,

As far as I know, stacking has a shared control plane (master has the control plane) and vPC has separate control planes. But what happens when you have a intervlan routing in your stack or vPC? Are all layer 3 packets routed by the master in a stack? Or do all members of the stack do routing?

For example, a virtual machine connected to a stack member that wants to communicate with a virtual machine on another VLAN but on the same stack member switch. Do packets have to go through to the master to be routed? or is there a local routing on the stack member?
As far as I know vPC does local routing

Thanks

[balaji.bandi]

Stacking used in all Catalyst enterprise switches, 1 Control plan for all the Stack member, MAster will have all the information interms of routing all other control plane related information. ( they are Logically 1) so 1 Gateway good enough to handle routing, in case master fails, all the infroamtion replicated to next master elected switch.

vPC - this more of the nexus side config used in DC Environment, here nexus work in Active / Actve mode

in this case if you looking Layer 3 to handle, you need to configure peer gateway and HSRP/FHRP to be deployed. Virtual IP wil be in Active node to handle, in case last Seconday node will take over, so Low distruption in services ?

is this make sense ? (i only touched based on your questioin)

[jmprats]

Yes, I know. Thanks

But, my question is, when stacking, all the routing information that the master has, is it replicated across the stack members so that they can route locally? Or should all Layer 3 packets traverse all the stack links from the stack members to the master and back to the members in order to be routed?

[balaji.bandi]

all the routing information that the master has, is it replicated across the stack members so that they can route locally? Or should all Layer 3 packets traverse all the stack links from the stack members to the master and back to the members in order to be routed?

Stack means Logically 1 switch, so only 1 brain, so all the information and logic will be with Master switch, rest of the members(switches) particiated in data only.

1 Routing table for all  - that controlled by Master.

High level all the routing process or control process take care by Master switch.

Hope this make sense ?

[jmprats]

Thank you very much. 

From your answer, I understand that all intervlan routing must pass through the master. Only the master performs layer 3 forwarding and the stack members just can do layer 2 forwarding.

But that can be quite inefficient. For me this could be the main reason to chose vPC over stacking in a datacenter.

[Giuseppe Larosa]

Hello @jmprats ,

I think that Nexus switches are also more powerful devices in hardware so the comparison is a little unfair .

Hope to help

Giuseppe

[balaji.bandi]

choosing the Switch is choice of design and Costing (nexus more meant to server or DC kind as mentione for fast switching)

Cat 9K onwards they are also more powerfull, as mentioned fast switching also in Stacking.

[Giuseppe Larosa]

Hello @jmprats ,

I would expect each stack member to have a local copy of the whole CEF table so that traffic is sent over the stack ring only when needed and the stack master is the destination only of packets that are punted to the main cpu for process switching.

At least for unicast traffic this can be done.

Hope to help

Giuseppe

[jmprats]

Thank you very much for your answer, any link with information about this? or how to configure it? Or is this the normal behaviour?

[Jon Marshall · Accepted solution]

You don't say which switches you are referring to but for the 9200 switches for example the master switch manages the control plane for L2 and L3 but the data or forwarding plane is implemented on all switch members (just as with modular switches). 

So L2 and L3 forwarding tables are pushed to each stack member - see this link for more details - 

https://www.cisco.com/c/en/us/products/collateral/switches/catalyst-9200-series-switches/nb-06-stackwise-architecture-cte-en.html#_Toc13521192

it will be the default behaviour. 

Jon

[Joseph W. Doherty]

As Jon notes, L2 and L3 forwarding is often both done by member switches.

In fact, on at least some switch stacks, they even support NSF (non-stop forwarding).  I.e. L3 forwarding continues even if the master switch crashes.