In large scale migrations DBAs are asked to make decisions with far-reaching and with unforeseen consequences. Deciding which approach to take is influenced by a number of factors including footprint size of inventory , duration of migration window , VMWare protection, application requirements , downtime requirements
Array-based replication
-replication happens at the storage layer
-replication level: LUN,VMFS or NFS
-Microsoft Cluster Services - MSCS cluster VMs can be replicated
-Downtime involved , coordination required
vSphere Replication
replication level :replicates at the VM level. More flexibility about moving the VMs
uses a different protocol than array based replication
downtime is minimal
MSCS cluster VMs cannot be replicated . vSphere replication cannot replicate disks in multiwriter , I've attempted to use this one previously - but
Physical RDMs are not supported in vSphere replication
The other common method is vmotion . The advantage of vmotion is minimal to no downtime . The disadvantage of vmotion is there isn't as much as flexibility - as opposed to vsphere replication
Ideally vMotion in less busy times, random access patterns on virtual disks interfere with sequential access of svMotion , causing negative impact.
svMotion migrate in 64kb chunks - this may have a negative impact on requests using a smaller size
Expect increased CPU consumption during svMotion .