Home > Storage > PowerStore > Virtualization and Cloud > VMware Horizon 8 with Dell EMC PowerEdge R6525 Servers and Dell EMC PowerStore 9000 > Virtualized desktop deployment strategies
Horizon 8 offers four choices for deploying virtual desktops: full-clone virtual machines, instant clone desktops, remote desktop session host pools, and manually provisioned desktops. In this document, two deployment strategies that are used in Horizon environments are discussed, focusing on their unique characteristics and their impact on I/O and CPU.
Full clones are independent copies of a virtual machine that operate separately from the original parent VM (the gold image) after creation. Any maintenance or updates of a full clone is done directly on the virtual machine itself, eliminating the need to access the original parent VM. When deployed to an automated desktop pool, Horizon manages the creation of full clones using a vCenter Server VM template or snapshot, creating a pool of identical VMs. Performance-wise, full clones function like a non-cloned virtual machine. Since it shares no virtual disks with the original parent VM, you can expect to see more optimal performance from these types of desktops compared with linked clones or instant clones.
Although PowerStore data reduction can help reduce the disk footprint, full clones require the most disk storage because they do not share a common base image. An important consideration with full clones is management overhead. Each full clone must be managed as an individual desktop and each full clone requires individual patch management and updating.
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 resulting in a persistent virtual experience. This VM type typically has the lowest overall storage I/O and CPU impact since data is saved between sessions. The VM is not refreshed at log off. The downside is management overhead as each full clone must be managed and updated individually like a physical desktop.
Activities such as patch management should be planned carefully as the impact on network bandwidth and storage can be significant. Leverage a local Windows software update server (WSUS) to avoid consuming Internet bandwidth needlessly. Maintenance windows should be defined and scheduled so WSUS updates a subset of machines that finish updating before the next group updates. This strategy helps to spread out the resource demands. Monthly updates can vary based on the number and type of updates that are released for that cycle, making it difficult to predict the resource impact.
Instant clones have replaced linked clones as the shared resource VDI provisioning method with Horizon 8. By using quick in-memory cloning of a running parent VM, and copy-on-write to quickly deploy the virtual machines, instant clone technology is the fastest provisioning method available in Horizon. Instant clones consume less storage and is a lighter workload for the vCenter compared to other provisioning methods available. Instant clones are I/O intensive during the initial phase of provisioning when ClonePrep files are created, and at user log off when VMs refresh from a copy of the current gold image.
A major advantage of instant clones is each machine is refreshed on every user log off, so it matches the current gold image. This behavior greatly simplifies the update process: update the gold image, and each VM is refreshed to use the new gold image at the next user log off. The gold image can be updated frequently without user interruption. The instant-clone deployment and update processes reduce management overhead and minimize end-user impact.
It is necessary to anticipate the significant burst of I/O and CPU demand on the storage array when using instant clones:
When user log offs are sporadic and spread out, the I/O and CPU bursts are spread out and less impactful. However, if a high percentage log off simultaneously, the burst in I/O and CPU demand stresses the storage array. A correctly sized environment can accommodate bursts in I/O and CPU due to pool creation and user log offs. The extra demand on the storage array from a CPU and disk I/O perspective must be anticipated and factored in when sizing the environment. The architecture in this document accounts for the bursts in I/O and CPU demand for these activities.
Dedicate a PowerStore array to VDI if the user count is high so that bursts caused by pool creation and image refreshes do not impact other workloads.