Kirill Kedrinski - Fotolia


Keep the cost of VDI storage under control

Layering, persona management tools and flash arrays help keep virtual desktop users happy and VDI storage costs down.

The costs of VDI storage historically have been so high they forced companies to compromise on performance or skip VDI altogether. Today, there are many options VDI shops can use to deploy the types of virtual desktops they want without losing out on performance.

In the early days of VDI, vendors sold the technology as a way to simplify management and save money. But soaring storage costs meant shops that wanted to do VDI had to find ways to minimize expenses. When it comes to VDI storage, cutting costs usually has an adverse effect on performance, which makes the user experience difficult and turns many away from VDI.

Since then, vendors have introduced numerous products to help address the performance/cost dilemma. They offer a wide range of options, including persona management, layering, data deduplication and flash arrays.

To persist or to pool?

IT shops looking to do VDI have two virtual desktop models to choose from: persistent and nonpersistent.

Many of the first VDI deployments followed a persistent desktop model in which a dedicated PC image is assigned to each user. Users can personalize their virtual desktops just like their physical counterparts. Whether they configure settings or install applications, those changes are available to them the next time they log onto their desktops. But the more users personalize their virtual desktops, the more storage they need. And as storage needs increase, virtual desktops become more expensive. They can even become more costly than providing each user with a physical PC. When shops don't add storage capacity for users' persistent desktops, the user experience deteriorates and productivity drops.

In an attempt to turn this situation around, many organizations look to nonpersistent desktops, which allocate stateless desktops from a shared pool. Users can't personalize nonpersistent desktops, but these desktops utilize resources far more efficiently, which means lower storage requirements and costs. For some use cases, the nonpersistent model works exceptionally well, such as in store kiosks or shared desktops in a classroom. But most enterprise workers prefer personalization, so companies get stuck between the persistent and nonpersistent models.

Bridging the virtual gap

When VDI and third-party vendors saw that neither persistent nor nonpersistent desktops were working out particularly well, they began coming up with ways to make both models easier, including persona management, layering and data deduplication.

Persona management tools provide a mechanism for saving user settings applied to nonpersistent desktops. For example, VMware offers View Persona Management to preserve user properties and synchronize them with a remote profile repository, which gives nonpersistent desktop users personalized profiles. It also adds a layer of complexity to VDI, and it requires additional support resources.

Desktop layering products from vendors such as Unidesk start with a single golden Windows image that can support both types of virtual desktops. IT administrators can patch and update the image as needed, and seamlessly apply it to either desktop type. All desktops share the same source OS and application layers, and admins can add a personalization layer to preserve user configuration settings without incurring lots of storage overhead. But, as is the case with persona managers, a layering product adds complexity and resource requirements to the VDI deployment.

Another option organizations have is data deduplication which removes redundancy in the VDI storage space. VDI tools often store a large number of images within a common storage space, and not surprisingly, the images share many blocks of the same data. Data deduplication helps optimize storage usage by eliminating the redundancy. The main issue with deduplication is its potential to affect performance; whether or not a dedupe product will hurt performance depends on the capabilities of the storage system itself, the amount of deduplicated data and the number of other services vying for system resources.

Storage issues unique to VDI

Boot storms. Because the majority of users log in at eight or nine in the morning and log out at five in the evening, the network gets overwhelmed with storage IOPS requests at those times.

Read/writes. VDI environments tend to have a high write-to-read ratio -- up to 60-70% writes -- which is much higher than the typical storage scenario. This puts even more strain on the system. An organization running thousands of virtual desktops would have to put up a substantial amount of cash to support such an environment, which brings up the next challenge:

Math. Estimate how much money each virtual desktop costs can be difficult because it depends on the storage system a VDI shop uses, as well as the structure of the VDI implementation. Estimates could range from $15 to $1000 per desktop, and several factors contribute to costs, including the amount of storage required per desktop, the number of virtual desktops, the amount of personalization on persistent desktops and the minimum performance requirements.

The flash revolution

Flash arrays promise to deliver unprecedented levels of performance compared to traditional hard disks. An all-flash array can deliver more IOPS to more virtual machines, which improves boot times and delivers better performance on all levels.

The big challenge with flash arrays is cost. All the extra performance comes with a hefty price tag -- five-to-10 times more than the typical SAN. As a result, some organizations have to load up each host with more desktops, which forces the system to handle greater I/O loads and offsets many of the gains that come with flash storage.

Some organizations turn to storage tools that take advantage of flash technologies without committing to all-flash arrays. For example, hybrid products create a single system for integrating traditional SANs with solid-state disks (SSDs).

Yet another approach is to use flash and SSD technologies for caching data, rather than storing it. This lets an organization reap many of the performance gains without making such a large investment, particularly if a SAN product is already sitting in the data center. The caching software keeps the most active data in the flash SSD and maintains the rest in the base storage system. Admins can also implement flash caching on the server side to reduce network traffic and optimize the caching process.

Next Steps

The latest layering tools help VDI

Complete guide to VDI storage options

How to get your VDI storage costs under control

Dig Deeper on Virtual desktop management