Home > Storage > PowerStore > Virtualization and Cloud > PowerStore: 1,500 VMware Horizon VDI users > Heavy writes
VDI environments are typically heavily write-based. After a VM is booted and the user logs in and begins work tasks, most of the I/O traffic consists of changes from the VM. The traffic consists of file changes, swap file writes, memory paging, and updates to user preferences.
During the steady state phase, the write ratio is typically 70% or higher. VDI has a higher write ratio than most other workloads. This write-heavy traffic pattern means VDI is a more demanding application for a storage array to support. PowerStore is well suited to provide optimal performance for the most demanding heavy-write VDI workloads.
Instant clones generate a large volume of traffic during the provisioning process due to the method used to create the VMs. Since the in-memory VM Fork technology creates the machines quickly, the volume of traffic is significant. The I/O required to complete the creation is brief but large. The VDI administrator must plan for bursts of high traffic on PowerStore when creating or refreshing pools of instant clone VMs. It is desirable to dedicate a PowerStore array to VDI if it supports a large VDI implementation so other workloads are not impacted.
Since instant clone machines reset on logout, as users log off, their virtual desktops refresh. Logoffs occur throughout the workday and the I/O and CPU demand to refresh desktops should be considered when sizing the environment. If many (or all) VDI users log off simultaneously, there is also a brief but significant burst in I/O and CPU demand as desktops are refreshed.
For this design, two instant clone pools are configured, with a dedicated Horizon Connection server for each pool. This configuration ensures load balancing across multiple connection servers. A randomization algorithm is applied to the login order to spread the login load evenly across the connection servers to ensure best performance and to better simulate a production environment.