Hyper-V cluster - one big CSV or one per host?

[rrbnc]

I have a 2-node hyperconverged Hyper-V/StarWind cluster. I have VMs running on both hosts. I believe StarWind V8 recommends configuring the storage as "fail over only" instead of "round robin", in order for the VM to make use of the local host's storage if possible. If this is correct, then should there be one CSV per host instead of just one big CSV shared between both hosts?

[anton (staff)]

It's always good to have at least one CSV per physical host with or w/o StarWind. So while it's technically possible to have one big CSV "community-owned" by many-many hosts that's not the best idea.

I have a 2-node hyperconverged Hyper-V/StarWind cluster. I have VMs running on both hosts. I believe StarWind V8 recommends configuring the storage as "fail over only" instead of "round robin", in order for the VM to make use of the local host's storage if possible. If this is correct, then should there be one CSV per host instead of just one big CSV shared between both hosts?

[rrbnc]

Thanks. Does it matter if the second StarWind datastore is under the same iSCSI target, or should it be separate? For example, right now I have Servers > HV1 > Storage1 > HAImage2. Should the new datastore be a second device under Storage1, or should there be a new Servers > HV1 > Storage2 > HAImage3 ?

[Anatoly (staff)]

Actually there is no difference from Starwind stand point

[barrysmoke]

planning for HA in the event of a node failure, it makes more sense to split a node's storage in 2 seperate targets, use 1 target for vm's that will run local, and the other target to replicate another node's vm's. then if one node fails, the remaining nodes that received the replication target, can take over the failed node's vm's.
you also get speed acceleration from a hyper-v node running vm's on a local target.

[Anatoly (staff)]

Agree, but that depends on the environment. For small deployments (<1TB) that doesn`t make significant difference, while adding complicity to the configuration.