Lately I have been mostly focusing my articles on VMware View 4.5 & Storage because in my opinion this is probably the single most critical component in any VDI solution, with networking in second. To finalise the series of storage related articles I decided to run an experiment to calculate how Linked Clones and Full Clones compare to each other in different scenarios. The objective of the experiment was to identify which scenario would require the least amount of storage disk space when all virtual desktops are consuming the maximum amount of storage available to them. An important first step was to create a baseline for comparison and I have set the following:
- Number of Desktops: 5000
- VM status: All Powered On
- Guest OS: Windows 7
- RAM: 2GB
- Parent VM Size: 30GB
- Parent VM Utilised: 8GB
- Refresh Linked Clones at: 100% (this will allow full utilisation of storage capacity)
- Persistent or User Data Disks are not utilised
- Disposable Disk size: 2GB
- VMs per Datastore: 64
- Snapshot: 1
- Memory Reservation: 50% (affect size of .vswp file)
To get the precise storage utilisation for each one of the scenarios I have also considered the following settings for each desktop:
- Size of the Linked Clone (Delta)
- Size of the Disposable Disk
- Size of the swap file (.vswp based on Memory Reservation)
The five scenarios I have specified as per table below.
No the interesting bits. The graph bellow represents the storage consumption for each of the scenarios above in Terabytes utilised. Read my analysis below.
The first thing to note in these results is that Dedicated Replica datastores really help to drive down the storage consumption, being for these scenarios something wound 6 TB. However, those dedicated replica datastores are accessed by hundreds or even thousands of virtual desktops at the same time and will require fast SSD disks and big caches to provide enough IOPS to support the workload. If you are not across Dedicated Replica Datastores read my previous article here.
The Linked Cloning scenario C is clearly the winner using only 35 TB; however we should take this result with a grain of salt. The scenario C does not include the disposable disks and all Windows swap IO write operations will be appended to the delta disk that may grow up to the total size of the Parent VM. This means that whilst there is storage savings you may need to refresh and/or re-compose those desktop more frequently. Desktops refresh and/or re-composed more frequently will hit storage more frequently, potentially requiring more IOPS during busy times.
Don’t assume that Disposable Disks are not useful. These scenarios were all calculate at full storage utilisation, but in real production environments you will have desktops over different storage consumption cycle and the disposable disk will help to reduce delta growth overtime.
I would love to be able to provide you with a IOPS study for these scenarios but unfortunately my 2TB Iomega is not enough for the simulation. I recommend you to stick with the recommendations from your storage vendor or run your own simulations.