We have a FAS-2240-1 running DOT 8.1.1 7 Mode HA Pair. One shelf fully populated and another DS4243 with 6 SSD drives, it serves both CIFS and FC LUNs. One aggr on each controller with 3 disks for root volumes. A second flash pool aggr with 21 disks. CIFS are shared our directly to the network using a 2 NIC VIF. All LUNs are presented to one 6 Node ESXi Cluster and another 2 Node ESXi Cluster. vSphere 5.1 supporting approximately 100 VDI Desktops and about 20 Servers.
As we start to grow the VDI environment it is time to re-adjust how we deploy the virtual desktop storage config, I think. Currently LUN config is as follows:
1. One for template systems. These are our parent VMs that get updated and snap shots taken to deploy linked clone pools.
2. One for the linked clones themselves.
When we deploy a linked clone pool the replica, windows swap files, and .vswp files are all on the same LUN. When we started our VDI deployment the numbers were low and after speaking with VMware they recommended that we keep everything together until we hit 75 or better linked clones, well we are there and getting the config correct for best performance and space usage has turnout to be some what of a challenge. All of our parent vm's have no reservations. Typical parent vm has a disk of 30GB thin provisioned and uses approximately 23GB, 3GB of RAM, 1vCPU with 2 Cores, and one NIC. When we deploy a linked clone pool the replica is approximately 23GB and the individual user desktops are approximately 4GB. In the end we will end up with about 10 linked clone pools.
I have spent a lot of time reading http://myvirtualcloud.net/?p=1421 , http://myvirtualcloud.net/?p=745 , http://myvirtualcloud.net/?p=1050 , and NetApp TR -3749 Version 4 plus more. It seems to me that config defiantly needs to change. As an FYI we use roaming profiles and full folder redirection to keep users data off the linked clones.
The plan is as follows:
1. One LUN for templates as we currently have. 10 parent vm templates 30GB templates so 300GB, de-dupe is on no snap shot space. Might be able to reduce it size once we can see de-dupe gains
2. Create one LUN for replicas. Just big enough to support the 10 replicas plus a few possible snap shots and re-composing. 10 replicas at 30GB each is 300GB plus 20% equals 360GB, , de-dupe is on no snap shot space. Might be able to reduce it size once we can see de-dupe gains.
3. Create one LUN for linked clones themselves. If I use the example size from above or 4GB and 100 VMs that would be 400GB. De-dupe is off and no snapshot space.
4. Create one LUN for linked clone disposable files. Tricky as some pools have different memory configuration, so if I go with the biggest one, 3GB, it would 3 times 100 so 300GB. De-dupe is off and no snapshot space.
Thats the plan. As always it is a point of departure as things change. Thoughts?