big VMFS datastores for shared VMs over iSCSI?

[mlubinski]

Hello,

I would like to ask you, what approach would you recommend if someone would like to build shared storage on iSCSI, but with as less datastores as possible?

I mean something like NFS volumes, where you can have one volume of 2TB and place there 100VMs or so. The thing is, that the less vmfs datastores will be visible - the better. So what I would expect is to have for instance 1.5TB vmfs, where I would be able to place 50 VMs with different sizes.

Are extents good to use for that purpose? What about if vmdk would have to be expanded from for instance 100GB to 300GB? (and corresponding LUN that would keep this vmdk would not have 250Gb free space for that)

Right now I am sticking to NFS, as it is very flexible, however i am also thinking about iSCSI - but then I can't predict changes on vmdk files.

If you found this or any other answer useful please consider the use of the Helpful or correct buttons to award points

[amvmware]

Why the issue of as few datastores as possible? - You can give them one datastore if you want but i would suggest performance of 50 or 100 VM's on the store will be poor.

The design of the datastores is to deliver the performance required and to limit or prevent scsi reservation issues.

For your requirements i think NFS is the way to go.

[mlubinski]

yeah, because it you plan to put few hundred VMs (with predicted capacity of ~4TB) on iscsi luns, then in order to follow up best practicies, you should go with ~400GB LUNs, so you would need to have 10 such vmfs datastores, and if you still are going to hit scsi reservation issues with 30+ VM's on one LUN.

i think I will stick to NFS design, as iscsi generates more problem with wide and unpredictable environments (like when you don't know if VM placed on 400GB datastore will suddenly not require to have 1.5TB disk attached

If you found this or any other answer useful please consider the use of the Helpful or correct buttons to award points