Home > Storage > PowerStore > Virtualization and Cloud > PowerStore: 3,600 VMware Horizon VDI Users > Windows I/O patterns
The evolution of Windows has brought new features to users but has created additional load for systems hosting it. From Windows 7 to Windows 10, the load increased dramatically. A larger portion of the increase was due to added services. There are other encryption and memory-protection schemes that are built into Windows. These serve to reduce the attack surface but also generate additional load.
As Windows 10 has evolved more security and features have been added that have increased the load even further. From the introduction to build 1909 the I/O generated by Windows 10 has doubled. This constant increase in I/O load requires matching increases storage performance.
The I/O patterns of VDI are a shift from the average read/write ratio of 70/30. The typical I/O pattern of VDI is 70-80% writes while users are logged in. Where the patterns differ between machine types is during the logout phase.
Instant clones perform a shutdown and refresh on every logout. This results in a new clone for each VM. This requires user data redirection to provide a persistent user experience. This VM type has the highest I/O load with the refresh cycles. It is very predictable as the load is directly proportional to user-log-out rates.
Linked clones use a common base image with a disk digest architecture to redirect changes. All changes to the VM are written to the digest. This can be a persistent architecture. If persistence is required, all changes are saved. Otherwise, all changes are flushed on logout. The disk load is higher in a non-persistent configuration as a new user profile gets created on each login.
The disk footprint of linked clones is very small. The base image is copied to each datastore in the pool and all machines share the base images. This greatly reduces duplication of the base image.
The I/O profile of full clones is that of full machines, with reads and writes spread across the entire dataset. Full clones typically save user data to the local profile. This results in a persistent virtual experience. This VM type typically has the lowest overall storage impact since data is saved between sessions. This is offset somewhat by the update requirements of full clones.
Since each VM is unique, it requires updates to be applied to each instance. The impact of thousands of machines updating can be significant and must be planned very carefully. Maintenance windows should be defined and groups of machines scheduled. The impact of each round of updates varies based on the updates that are released. This makes it difficult to predict the impact.