BACKGROUND IMAGE: iSTOCK/GETTY IMAGES
An increasing number of enterprises are using desktop virtualization to reduce their cost of supporting PCs. But while a virtual desktop infrastructure provides the benefits of centralized applications, it also changes how individual users are supported. If your infrastructure is not suited for the VDI model, performance and stability issues can be profound -- and potentially disastrous.
In a virtual desktop infrastructure (VDI), applications run on a server virtual machine (VM) and are linked to the user via a desktop client app. The actual application execution is remote, and PC storage, memory, and CPU access are virtualized and hosted. The master virtual PC can be projected to any suitable client device, but only the user interface is projected. Since the desktop instances are hosted, they can be affected by the same problems as server virtualization applications -- and more.
The big challenge when planning for VDI hosting is the sheer number of virtual desktops. Most companies using server virtualization run two to five virtual servers per actual server: A large enterprise might host 2,000 to 3,000 virtual servers. However, that same company could have 10,000 or more desktop systems to virtualize. Predicting how all those virtual desktops will use the data center resource pool is a challenge.
Instead of treating PCs as discrete systems with their own operating systems and middleware, software and storage, virtual desktop technology lets enterprises create machine images of various "classes" of systems and load those images on demand. In some cases, users can customize the configuration of the master image in the same way they would customize the configuration of a real system. But customization means more desktop master images to manage and changes to application requirements can make an old master incompatible with a worker's current usage.
In terms of resources, memory may be the toughest VDI issue to manage. Unlike server application components that can have short persistence -- particularly in service-oriented architecture software -- desktop applications are designed to load and stay running for hours: They must be paged out to be removed from memory. That paging can create nonsustainable disk I/O loading. Even if a given set of users run the same basic application, in most cases, they can't run the same exact copy. Therefore, a large memory pool that can hold as many discrete machine images as possible is essential.
Disk storage is another challenge with hosted VDI applications. On real distributed desktops, client system disk usage is supported on different devices and controllers, and thus, it would never collide. When desktops are virtualized and hosted, the host systems has to field disk I/O for all of the virtual desktops at the same time, which can create congestion and performance problems, particularly if work schedules can produce frequent synchronized behaviors. If every user starts his/her day by reviewing a task list, the 9 a.m. I/O impact can be profound. Therefore, it's critical to have very efficient I/O and storage systems on all VDI hosts.
Affordable solid-state drives are an advance that impacts both memory and storage. Solid-state disks and effective multilayer managed caching of machine images and paging can reduce the memory requirements for a given level of application performance.
Multicourse server technology is also an enhancement to VDI support. Remember the total CPU power of 10,000 desktops was available to support the hypothetical enterprise in a standard client/server mode. Compressing those desktops into a set of VM resources is more likely to succeed if every server has several cores where application needs can be allocated to. Otherwise, a collision of activity could reduce performance to near-zero levels for all.
The biggest infrastructure challenge for the hosted virtual desktop model is sustaining the performance of the server-to-user connection. Unlike client-server computing, which exchanges basic data elements between the desktop and the server farm, virtual desktop computing must provide a remote display and keyboard interface that can be significantly more bandwidth intensive. Since the performance of the communication's connection is critical to user satisfaction, VDI management plans have to take the capacity of this link into account. When the desktop and server are in the same physical facility, only LAN capacity is consumed, and companies can improve virtual desktop performance by increasing the speed of their LAN connections (both to the user and between LAN switches). Enterprises can also flatten their LAN infrastructures to reduce the number of LAN switches between real desktop users and virtual desktop host systems.
Many companies are now considering or deploying virtualization and cloud computing, and in the process, they're refining their data center networking. This is a good time to consider and address the network impact on VDI performance. Flattening the data center and headquarters network improves virtualization and cloud computing performance in the data center as well as VDI application performance.
In cases where VDI supports remote workers, performance will normally be linked to the capacity of the remote access connection. The explosion in consumer broadband has made "business Internet" services with access speeds of 10, 20, 50 or even 100 Mbps available at reasonable costs. Using a VPN with such a service may be the best way to ensure good application performance for remote VDI users.
VDI technology is justified by operational savings, but those savings cannot be realized if business operations are disrupted by performance issues. Invest in adequate VDI resources upfront to properly support your enterprise. As always, conduct a limited-scope pilot test to verify the conclusions of an infrastructure assessment. With careful planning, a VDI project can significantly reduce current costs and contain further cost increases associated with growing PC support demands.
ABOUT THE AUTHOR:
Tom Nolle is president of CIMI Corp., a strategic consulting firm specializing in telecommunications and data communications since 1982. Nolle is a member of the IEEE, ACM, Telemanagement Forum and the IPsphere Forum, and he is the publisher of Netwatcher, a journal in advanced telecommunications strategy issues.