Behavior during VPC enablement

[jbekk]

I'll try keep this simple using pictures... We want to get from this:

 To this:

The basics are:

• We are consolidating DC locations for this company/branch as part of a bigger picture strategy.
• No redundancy with switches at either of these sites exists currently
• We want to relocate and reuse the core switch from DC2 to provide core switch redundancy at DC1.
• All in production servers and sites are exclusively connected into DC1-CORE1 (i.e. consolidation for those parts of the environment have been completed already).
• We are planning on adding the VPC feature to both DC1-CORE1 and DC1-CORE2.

Now, provided nothing is connected to DC1-CORE2, I don't believe any disruption to traffic forwarding/routing would occur on DC1-CORE1 or its connected workloads. This is what I am after advice on... has anyone had experience enabling VPC on an in-production environment before?

NOTE: After the second core is added and VPC is enabled, the plan is to start redundantly connecting Layer 3 links and port-channels/etc to both switches.

[ADP_89]

Hello,

Adding VPC won't create any problems as long the configurations are consistent(STP, vlans etc... https://www.cisco.com/c/m/en_us/techdoc/dc/reference/cli/nxos/commands/vpc/show-vpc-consistency-parameters.html) and the operations are done in the right order. Just make sure that the DC1-CORE1 will be the primary by setting the role priority to a lower value.

I would proceed in the following way:

- Cable up CORE1 to CORE2 (peer-link and keepalive)

- Check if the links are going up, you can create the port-channel and not trunk any vlans on it just to check layer1 and lacp are working.

- Shut down the peer-link on DC1-CORE1

- Configure VPC on both sides, add the "  vpc peer-link" on the Po as well

- Reload DC1-CORE2, while it reloads unshut the peer-link on DC1-CORE1.

- Wait DC1-CORE2 to come back up and you should have a VPC pair up and working

The reason why I would reload the box is to avoid the VPC sticky bit clash in case you will wait to much time before bringing up the peer-link(that would cause disruption).

HTH,

ADP

[jbekk]

I tend to agree with this. It should be fine provided everything comes up cleanly and roles don't jump about/etc. I think I am going to have to lean on TAC here and get something firm though to manage the risk on this one. I'll come back once I hear more.

[nazimkha]

Hi,

Before you even start implementing there are lot of things which you may need to plan ahead in the designing part.

Going with vPC has it's own benefits but you may hit into problems if it is not designed correctly, for instance even if you have both cores DC-1 and DC_2 in vPC but you do not have any dual homed devices than vPC would be an extra overhead without any benefits.

You mentioned connecting layer-3 links and layer-2 links to both cores....but you have not specified the distance between the DCs. You may also have to consider the BUM (Broadcast, Unicast and Multicast) traffic. Where would be your default gateway's, how will be the traffic flow.
You can go through the cisco live presentation on vPC best practices. Link to my session is below
https://ciscolive.cisco.com/on-demand-library/?search=vPC#/session/14479207929320017eHp

[jbekk]

You are trying to provide advice on whether the VPC feature set is suitable and/or viable rather than provide guidance on whether the switch will drop Layer 2 traffic when the VPC feature is enabled. You have also started to provide guidance as to how two switches in different DCs aren't really suitable for doing VPC with one another... and completely missed that I point out that we are physically relocating the switch from DC2 into DC1 (thus the distance between DCs is not a consideration/concern).

Sorry but this is not what I asked for advice on... Thanks though.

[nazimkha]

Oh My Bad. I didnt read through it properly. I was more of thinking vPC as a DCI.

Adding a vPC switch should not create any issues if the configurations are consistent but you may never know if some vpc vlan may become inconsistent