StoreVirtual Storage
Showing results for 
Search instead for 
Did you mean: 

iSCSI Switch replacement - best practice suggestions needed

Paul Miano
Occasional Advisor

iSCSI Switch replacement - best practice suggestions needed

We are looking to replace our Cisco 3560G switches with 3750X models. We are using ALB and have a layer 2 link via an SFP (Single Mode FIber). One option for the swap is to scheudule and outage for the SAN volumes, but I'd rather figure out a way to do it 'hot'. Has anyone tried this? Any suggestions? Warnings?


We are preparing to go to a multi site (single subnet) configuration and we are upgrading to teh 3750X so we can have 10G uplinks between the switches, btw. For what it's worth, we are running only ESXi hosts connected to the SAN and we have 4 nodes in the SAN cluster (2x 2120, 2x 2120 G2).

Bryan McMullan
Trusted Contributor

Re: iSCSI Switch replacement - best practice suggestions needed

While you could do it while on-line by being very methodical and moving one connection per unit, then one unit at a time (as long as both the old and the new switches have a high bandwidth link between them).  I'd personally think you could get it done a heck of a lot faster by just downing your systems for and doing the move wholesale at one time.  Overall, it's safer and would prevent unplanned outages when something just isn't configured right.


So, an on-line move is possible and shouldn't be too hard, I don't think I'd risk it personally.

Kirk Woloshyn
Occasional Contributor

Re: iSCSI Switch replacement - best practice suggestions needed

I've done this on both of my SANs within the last couple of weeks to accommodate growth. Since all of my servers use MPIO and are connected to two switches, I put a third switch in place, cross connected to one of the old switches and moved the cables from the old to the new one node at a time. I always waited 1-2 minutes between moving nodes to ensure the node I just moved was reconnected to the cluster. If you move the cable quickly the node won't even throw an error.


Just go slow, watch for alerts and make sure that if you get one it clears before moving on. It took me 15 minutes to move six nodes and four servers from one switch to another. Users didn't know it happened.