How does deduplication help workloads on virtual desktops? Are there any limitations to VDI deduplication?
Endpoint virtualization that uses technologies like virtual desktop infrastructure (VDI) has gained attention in recent years as organizations seek to centralize endpoints and impose security while reducing hardware dependence. But storage has been an important limitation for VDI deployments because each endpoint is essentially deployed as a unique virtual machine (VM). It makes sense that data deduplication would be a noteworthy benefit for VDI because significant storage reductions could vastly expand the number of desktop images hosted on each server while lowering the number of servers required for enterprisewide VDI deployments.
Traditional deduplication didn't work well for VDI because of its open/active file behavior. Windows Server 2012 R2 specifically addressed this problem by adding support for VDI VMs. Deduplication is still implemented as a periodic background job, so even though VDI files may change constantly, portions of each virtual hard-disk file older than the number of days specified for the configuration will be deduplicated.
There are requirements that support VDI deduplication. First, the VDI instances must be stored on a system other than the Hyper-V host supporting the instances. This isolation is important to prevent deduplication processes from impairing the VDI machine performance. This does not require a change in the way that Hyper-V or VDI is used, but the VDI instances simply must be stored on a different Windows Server 2012 R2 file server. Second, storage must be connected using the Server Message Block 3.0 protocol, and technologies like Scale-Out File Server or Cluster Shared Volumes are recommended additions for scalability and high availability in the VDI deployment.
However, the high levels of compression possible with deduplication also make it possible to deploy a high-performance storage tier such as solid-state disk drives -- combining efficient storage with high-I/O levels for better VM performance across every virtualized endpoint.
This was first published in February 2014