I have a need to create a VM that needs multiple terabytes of space on a single drive letter. I've read in the past to avoid using extents to expand the datastores but it seems that this was with vmfs 1 and 2. What's the consensus on using extents in vmfs3?
Is using extents in vmware still a bad idea?
- Thread starter Dr. Xing
- Start date
- Status
I'd use RDM instead. If its FC storage, you can also use NPIV if your switch/hardware will support it and talk direct to the array utilizing WWWN etc.
the real danger with extents is when you have a given vm that ends up sitting on multiple LUNS. If you lose an extent then VM's that are on the other extents of a datastore should be ok. But if a VM spans two extents for some reason, it's cooked. Also, if you lose the first LUN in the VMFS volume then you lose all of the other volumes as well.
But how often do you have failures that lead to bringing a LUN down?
But how often do you have failures that lead to bringing a LUN down?
The VMFS limit of 2TB-512B applies to RDMs as well. Bummer, eh? 
Since you need the guest OS to see a single volume that is >2TB, I think that an extent would not help you anyway since with the maximum block size of 8MB, you'd still be limited to a max file size (i.e. a single VMDK) of ~2TB...
Since your only option is DAS, one way to accomplish your goal (albeit not totally pretty) could be to create several datastores each with a single large VMDK, attach all of these VMDKs to the one VM, and place all of the disks into a software RAID within the guest OS.
With iSCSI, you could create a LUN and present it directly to the guest OS's iSCSI initiator. Also, since the limitation belongs to VMFS, I believe that a datastore backed by an NFS export would also be a solution to having one very large VMDK.
EDIT:
Here's an article on the VMware Community where the reason for the 2TB limit (and why it applies even to RDMs) is explained.
Since you need the guest OS to see a single volume that is >2TB, I think that an extent would not help you anyway since with the maximum block size of 8MB, you'd still be limited to a max file size (i.e. a single VMDK) of ~2TB...Since your only option is DAS, one way to accomplish your goal (albeit not totally pretty) could be to create several datastores each with a single large VMDK, attach all of these VMDKs to the one VM, and place all of the disks into a software RAID within the guest OS.
With iSCSI, you could create a LUN and present it directly to the guest OS's iSCSI initiator. Also, since the limitation belongs to VMFS, I believe that a datastore backed by an NFS export would also be a solution to having one very large VMDK.
EDIT:
Here's an article on the VMware Community where the reason for the 2TB limit (and why it applies even to RDMs) is explained.
I've got a customer running a bunch of 16TB linux file systems in vmware.... Basically it's 4 P2000's with 4 1.3TB Lun's each with a 1TB VMDK. All 16 1TB vmdk's are presented to a VM and software RAID stripes across them. Running IOzone benchmarks of various settings we have hit 2GB/s throughput on the systems. Managing the VM's is a pain and I the VM's have 100+GB of RAM. I tried to get them to not bother with making them VM's but they insisted...
It works and seems stable.... but still it worries me on a consistent basis....
I still say large fileserves are a candidate for not virtualizing....
It works and seems stable.... but still it worries me on a consistent basis....
I still say large fileserves are a candidate for not virtualizing....
- Status