We're using our FC SAN primarily to house data used by database clusters. These databases (usually VM's) need storage that understands the SCSI-3 reservation protocol so clustering will function properly. We expose the LUNs to VMware and map to the guests as RDM's. This works quite well and delivers low-latency performance for environments that require the functionality clustering provides. We can't currently deliver this via our primarily NFS-based datastores (VMDK on NFS doesn't support SCSI-3) or host-based VMDK's (SCSI-3 will work if your VM's are both on the same hosts, but we want to have cluster members on separate ESX hosts for obvious availability reasons).
My question is whether the newly announced "Virtual SAN" can deliver SCSI-3 support without my traditional/legacy FC-based SAN? This obviously could save a lot of money, but in reading through all the blog posts and announcements, it's not really clear to me if handling MSCS/GFS2 type workloads is or isn't part of what it can deliver. Obviously VMFS itself is a cluster file system, but I could see the latencies associated with some of the reservation activities as things bounce around between ESX hosts over Ethernet (IP?) hurting performance and make me miss my fiber channel setup.
Anyone have insight on this? Perhaps we should be architecting things differently and getting away from traditional clusters with shared LUNs... even VMware FT doesn't quite give me the protections I need though.
Thanks!
Ray