Storage Considerations for Virtual Desktops: Sizing your VDI storage to ensure optimal performance

Welcome back folks. This is the second post in a series of three blog posts discussing storage considerations for VDI deployments. In my last post [BV1] I highlighted the critical success factors for your VDI projects, and how your storage infrastructure plays a central role in ensuring successful VDI roll-out. I also mentioned that one way to meet the desired VDI performance is to increase the number of disks in your storage environment. While this can lead to over provisioning of storage, it is important to understand how increasing the number of disks can help you meet your performance goals. In this post I will provide a simple example that illustrates how to size the storage for your VDI environment in order to ensure optimal performance and user experience.

Sizing storage for VDI can be especially tricky. On one hand, you need to ensure realistic performance. On the other hand, you are under a tremendous pressure to minimize storage footprint in order to lower storage related costs. When designing a VDI environment, it is critical to ensure that your storage does not become a bottleneck. VDI is a highly transactional workload and its performance is bound by disk I/Os. One way storage administrators can meet the I/O demand is by increasing the number of storage disks supporting their VDI environment. It is important to look at the peak IOPS demand that will be placed on the underlying storage, and design your storage to handle those peak IOPS. There is no simple formula for assessing your IOPS needs. It will very for every organization depending on their desktop user types, and boot and login patterns. I will take a simple example here and go about identifying the peak IOPS needs and further the size of the storage needed to support those IOPS.

VDI Storage Sizing Exercise

Let us say that we are designing a VDI environment for 100 desktops. We estimate that about 80% of those desktops will boot in the morning simultaneously and then stay online rest of the day. Remaining 20% of the desktops will boot and perform operations during the night time.

Let us assume that we estimate each desktop to produce a peak of 25 IOPS while booting up, and then an average of about 5 IOPS throughout the steady state operations before logging off. The logoff operation creates a peak of 15 IOPS per desktop.

We also estimate that we will grow our VDI environment by up to 50% in the next 5 years and we are designing the VDI environment to support this growth.

As it is evident, the boot operation is the most I/O intensive and thus, we should be sizing our storage to support the performance requirements during boot periods.

So the 80 desktops (80% of 100 desktops) will produce 80 X 25 = 2000 IOPS when they are booting up in the morning simultaneously.

To account for the anticipated growth, we will need to support up to 50% more IOPS in the next 5 years as we grow our VDI environment. So we need to design our storage to support 2000 X 1.5 = 3000 IOPS in order to effectively address the peak IOPS requirements for the next 5 years.

Let us assume that we will be using 15K RPM SAS drives to support this VDI environment, and that each 15K SAS drive can support 150 IOPS.

Hence we will need 3000 / 150 = 20 of the 15K SAS drives in our storage to handle the peak IOPS.

This example assumes RAID 0 policy and no spare disks.

This is a simple example, but should help illustrate the decision process. The key is to estimate, as accurately as possible, the IOPS patterns for your specific environment. It is important to note that the IOPS patterns vary drastically between organizations. Dell offers consulting services that can help you analyze your VDI needs to meet the required performance goals.

Take Away:

Inadequate storage performance in your VDI can lead to less than acceptable user experience. While sizing storage for your VDI, it is important to design it to deliver adequate performance and user experience. VDI storage needs to be sized to handle peak IOPS demand placed by the VDI workloads. The I/O patterns vary drastically from organization to organization, and a critical analysis of your end user profiles is essential to accurately estimate the peak demands placed on your storage infrastructure.

It is important to recognize that overprovisioning your storage in order to meet the performance goals is cost-inefficient. In my next and the last post in this series, I will discuss some critical features and considerations for your VDI storage that can help mitigate performance challenges while optimizing storage footprint and enabling a simple, streamlined desktop VM provisioning model. Stay tuned and be sure to stay connected with the latest updates by following @Dell_Storage.

About the Author: Vikram Belapurkar