Migration to new network switches - takeover vs migration

[ajs]

We are migrating our two-node FAS2552 (9.3) cluster to new switches. The new and old switches have L2 connectivity. All 4 10G ports are in use on each node. There are 2 available 1G ports on each node. I've read forum posts that describe, broadly, two options:

1) use takeover to move all workloads on to one node, connect the freed up node to the new switches, giveback, rinse and repeat with the other node, or

2) migrate the data and LIFs so that they are all on one node, connect the freed up node to the new switches, migrate workloads back, rinse and repeat with the other node.

The downside to #1 is (my understanding) that when you move the node to the new switches, you cannot verify network network connectivity before doing the giveback - so you need to be sure the networking is correct.

It seems the only downside to #2 is that it is more typing/work and perhaps that means greater potential for error.

I can model #2 using the available 1G ports on the old and new switches with some test volumes and LIFs.

Of course, the goal is to do this non-destructively, except for any CIFS session blips.

What do you think would be best? Are there other options to consider? Thanks!

[SpindleNinja · Accepted solution]

I typically do (and prefer) opt 2 as you have more control.    

[CGulde]

There is a detailed procedure for replacing Cluster switches which doesn't involve moving load or LIFs.

The procedure below is for migrating form CN1610 to NX5596 switches but can be adapted to other switches.

https://library.netapp.com/ecm/ecm_download_file/ECMP1140532 

[ajs]

Starting to flesh out my plan for testing. Is there a way to move the storage from one node to another? Or will only be able to migrate the LIFs to the other node (and in this case, I presume the storage traffic will cross the intra-cluster network)?

[aborzenkov]

You can move volume to any aggregate in a cluster and you can relocate aggregate ownership between nodes in HA.

[ajs]

I've never relocated aggr ownership in a cluster. Does this process take a while, like a volume move would, or quickly, like migrating a LIF?

[SpindleNinja]

I wouldn't even worry about the aggr relocate for this, or even volumes. 

The LIF is the more important part,  even if you move a volume to the other node, and not the LIF,  you will still be sending data out where the LIF is.     If you do have a 1-1 volume to lif relationship for your config, you can move both if you have the capacity on the other aggrs.  

[ajs]

Thanks all. Our migration went off without a hitch (well, we did lose a ping or two along the way). We moved the LIFs to the non-home node, moved the node to the new network, and then migrated the LIFs back. And likewise for the other node. Worked well. Thanks for the input.

[SpindleNinja]

Good to hear!   To note, a dropped ping as the lifs are moved is normal.