Excellent comments, Glenn. I appreciate your perspective.
- You have to (by default) allocate 2x + deltas for LUNs. For a
100GB LUN, you'd need 240GB of space (by default--I know there are workarounds with ONTAP 7.2 and later). With an NFS mount (which is nothing more than exporting a FlexVol), you only need to account for the 20% Snapshot reserve...
Be careful! The reason you need 2x + deltas is to simply ensure that you maintain enough space in the volume to guarantee that you'll be able to overwrite data if you decide to overwrite the entire LUN (say, if you format the drive while a snapshot was there, or maybe re-indexed or defragged the NTFS volume in a VMFS filesystem?) Just because you choose a 20% snap reserve in a NAS volume doesn't mean that you won't blow through that in this case in NFS, and end up with the same problem with "out of space"- you'll do a DF and see that your snapshot usage is like 1000% while your regular fs usage is only 40% or something like that. And since netapp is all about preserving data in all cases, you won't be able to do anything until you delete snaps. Bottom line is that you don't get a totally free lunch with NFS, although the thin provisioning DEFINITELY helps. Don't forget flexclone, if you're using VMWare and NOT using Flexclone you're missing out.
Also, if you don't do snaps, you don't need the size of 2x + deltas, but then again why would anyone do that?
Right--I understand the need for the fractional reserve and snap reserve, and I understand that snap reserve in a NAS volume may well exceed 20% and eat into the rest of the space for the FlexVol. It just seems much easier to manage, overall, with NFS than with iSCSI and LUNs, IMHO.
- You can't resize iSCSI LUNs. With NFS on a FlexVol, you can
resize to your heart's content because WAFL is controlling the filesystem--not the host.
Sure you can! The problem is that VMWare doesn't like it when you do. Windows and Linux couldn't care less, they see more space to extend their partitions. VMWare hasn't gotten there yet.....I don't think. I know it blew up ESX 2.5.
As Peter pointed out, you *can* (but that doesn't necessarily mean that you should) resize LUNs that are formatted as VMFS. ESX 3.x handles it much better than 2.5 did, but there is some debate as to whether extents should be used or not. Again, having WAFL control the filesystem means that the storage system is back in control, and we remove some of the limitations placed upon us by the host OS.
- It's open, meaning that your VMDKs aren't locked into the
proprietary VMFS file system. This could potentially simplify backups and restores.
Totally agree, good point. One other thing you need to consider is that you'd still likely keep each VMWare Guest VM in their own volume and export each one, to be granular with snapshots + replication. It's more work but well worth it.
Right, this is a design decision that will apply whether you are using NFS or iSCSI. You will need to be careful, however, that you balance your deployment--there is a limit on the maximum number of VMFS datastores that can be managed by a single VirtualCenter.
Again, great points, Glenn!
Scott Lowe ePlus Technology, Inc. slowe@eplus.com