Replacing Access Layer Switches in a FlexPod design - Do we need to shut down the VMs?

[klopez138]

Hello,

            We are in the planning phase for a maintenance window to upgrade the access layer switches in a design very similar to the one below. I'm curious as to whether or not we need to shut down the VMs that are running on the blade server before we move the iscsi connections. It's my understanding that when an iscsi cconnection is interrupted, reads and writes are suspended to ensure everything stays intact. So, I was thinking that we'd be able to rack in the new switches, quickly move all the iscsi connections from the old switch to the new, and cutover the remaining mgmt and other vmnic connections after all the iscsi ones have been moved.

             I wanted to see if anyone out there who does maintenance like this more regularly may have some best practices they may want to share. Full disclosure, I'm not a datacenter guy and have never swapped out the access layer switch in production in a design like this. We've had switches like this go down before without having issues with the VMs which is why I was thinking it would be ok to quickly move the connections from the old switch to the new one.

[balaji.bandi · Accepted solution]

is this upgrade or replacemnt. ?

if upgrade - since all layer 2 all build vpc, i will build new vpc on new switches to connect to parent switches and do cut over one side at a time to  minimise the downtime, if the business agree full downtime, big bang migration and test it.

[klopez138]

It's a replacement of the access layer switches. So there is no VPC to be created on the new switches, only the port channels that connect to the vpc on the Nexus switches.

[klopez138]

It's a replacement of the access layer switches. So there is no VPC to be created on the new switches, only the port channels that connect to the vpc on the Nexus switches.

[balaji.bandi]

is this exiting one vPC , so if you like take advantage of vPC , i would suggest vPC all nexus device where possible, until you see any Limitations of the design,. some vendor may not support or take control of whole path outage.