By replacing physical computers on employees’ desks with a virtual computer running in a hypervisor in the data center, desktop virtualization, like server virtualization, leverages consolidation of the compute infrastructure to generate savings. But the benefits of desktop virtualization don’t come from lower hardware and software costs, as is typical in a virtualized server environment. Instead, the reduction of management overhead...
required to keep a collection of desktop computers provisioned and updated drives the cost savings. In fact, consolidating desktop storage is one of the reasons virtual desktop infrastructure (VDI) implementations often end up costing more than anticipated.
Most desktop computers come with a fairly large hard drive, up to a terabyte in many cases. Most desktop users don’t come close to filling up their local hard drives, but even at 50% of maximum capacity, in a VDI environment, each desktop could add 500 GB to the consolidated storage requirements. Desktop or laptop hard drives are typically the most economical, capacity-centric drives available, which is adequate for the applications most users run. But the enterprise-grade, shared storage that’s typically used to support a VDI hypervisor can easily be 10 times the price of internal desktop drives. And, replacing existing PC-attached disk drives essentially means buying that capacity a second time.
Performance requirements of VDI storage have an even greater role in driving up VDI cost. Just like in server environments, virtualizing desktops creates a consolidation of resource demands, but on an even bigger scale since more virtual desktops can be put onto a single hypervisor than can server VMs. The spikes in storage requests from intermittent events can cause IT to overallocate storage resources in an effort to maintain consistent performance under load. Examples can be “boot storms,” when most users log on in the morning; usage tied to regular events like monthly financial activities; or automated virus scans. The solution to cover these increased but temporary I/O requirements can be to buy more or faster storage.
Unfortunately, throwing money at the problem isn’t an option for many (most?) IT shops, so managers will have to find out how to address these issues without breaking the budget. Whether it’s keeping up with capacity or justifying a new VDI deployment, ways to control storage costs is something most of your customers need to understand.
Storage efficiency techniques
One strategy to help keep VDI costs related to storage under control is to leverage technologies developed for the storage industry at large, essentially making the storage that’s implemented as efficient as possible. Those technologies are:
Thin provisioning. Thin provisioning has established itself in the storage world. By keeping track of how much logical capacity has been allocated (or promised) to each VM and then physically allocating only as much as is needed, the amount of wasted capacity in an environment can be reduced significantly. Since desktop users’ capacity requirements are relatively predictable, thin provisioning is a good technology fit for VDI.
Golden masters and linked clones. Since users can have similar compute requirements, the images that make up each virtual desktop computer are also similar. To leverage this similarity, a golden master can be used; this is an image file that includes the core OS, applications and common files that every desktop will need. But instead of copying these golden masters for each new desktop, a linked clone uses pointers to the master image to create a file that is much smaller than a full copy would be. The result is a significant reduction in storage consumption.
Deduplication. Deduplication was originally developed for backup systems but has more recently made its way into primary storage. While the data reduction ratios aren’t as dramatic in VDI as in backup, it’s still a worthwhile step to take. Since dedupe uses pointers to reference a single instance of a file or subfile block, it’s well suited for VDI, where users typically create similar data sets or keep copies of the exact same files.
Lightening the storage processing load
Reducing cost isn’t as simple as implementing the storage efficiency methods described above. IT shops must also consider the hypervisor. Most of these efficiency processes require dynamic allocation of storage, something that the hypervisor is not particularly good at. The result can be a significant impact on performance. For many environments, the only option is to move these storage services onto the storage array, where storage controllers have the processing power to handle this real-time load efficiently. Unfortunately, this can mean buying storage systems that include these advanced functions and paying even higher costs per gigabyte on storage that was already more expensive than the capacity it was replacing.
But your customers do have options that don’t require additional hardware. One of them is a software product from Virsto that installs on the hypervisor and sets up a write log type of architecture to maintain performance while it improves the efficiency of the storage connected to the hypervisor. This solution temporarily holds writes in a higher-performing storage area (fast disk or solid-state storage) while providing acknowledgement to the hypervisor (the VM making the write). Writes from the log to disk can also be made in a more efficient manner, more sequentially and logically grouped, so that read performance is enhanced as well.
Virsto isolates the requesting VM from the latency of back-end storage, similar to the way a transaction log manages writes to a database. Using a purpose-built, log-structured file system, the software manages storage for each virtual desktop, including storage services such as snapshots, clones and thinly provisioned volumes. This essentially removes the overhead from the hypervisor, much like putting these services onto the storage array would.
Another benefit of Virsto is its ability to use high-speed disk, flash or even battery-backed DRAM, for the write log area. This can further improve overall performance by providing faster response to the hypervisor, but it can also provide a high-speed storage area for the golden images used to create linked clones.
Most importantly, this type of technology can save storage costs. By eliminating the impact of storage latency on the requesting VMs and making disk writes more efficient, this technology can eliminate the need to buy high-end storage systems in order to maintain consistent VDI performance.
For VARs, the message from this tip is opportunity. Since VDI cost issues can be counterintuitive and the solutions complex, there should be considerable interest for these kinds of strategies. The traditional “solutions” for storage performance -- adding more disk drives -- may only be marginally effective and will only make the cost issue worse. A VAR who understands the real causes of these problems can propose solutions like that from Virsto or storage efficiency methods to help users postpone a storage upgrade. This kind of thinking can differentiate a VAR from the competition and demonstrate its value to a new potential customer.
Eric Slack is a senior analyst with Storage Switzerland.