|
Like other virtualization products, desktop virtualization holds the promise of dramatically simplified IT infrastructure, cost effective IT utilization and management, with greater security around enterprise data. Vendors such as VMware, Microsoft, and Citrix claim that they can provide enterprises with a new model of enterprise computing, whereby any thin client can be used to securely access company data without the need for an individual to have their own machine, with their own data, and their own applications running on it.
As with most distributed computing models, desktop virtualization is faced with a serious performance challenge when enterprises attempt to use this technology to support distributed users who may be located across the country or across the globe. While some desktop virtualization products attempt to use communication protocols that are designed for wide area networks (WANs), even these products run into two fundamental challenges: bandwidth constraints often limit the amount of data or the number of users who can access virtualized desktops, and latency prevents the users’ applications from having local-like performance.
Wide-area data services (WDS) is a class of technology that has rapidly been adopted across enterprises in order to deal with the challenges of bandwidth limitations and latency over the WAN and enabling LAN-like performance for remote users. While WDS has had tremendous success in enabling server virtualization and remote site consolidation, typically WDS has not been deployed to support desktop virtualization environments. This paper will explore WDS in more detail as it relates to different desktop virtualization models. Three different architectures will be presented for integrating WDS and desktop virtualization. In addition, performance numbers for each architecture will be presented that show acceleration by WDS solutions.
Architecture #1: Centralized virtualization gateways
Architecture #1 (shown in Figure 1) represents the most common deployment style of desktop virtualization today. Enterprises deploy a virtualization technology such as Citrix Presentation Server or VMware VDI in their data centers, essentially sitting in front of the applications that are required by users in the headquarters or remote offices. Those offices are equipped with thin clients or desktop machines that must connect to the virtualization server in order to load a desktop or access any application. In this model, application data is not traveling over the WAN. The virtualization technology is transmitting the visual representation of the desktop along with changes in the desktop – mouse movements, highlights, or new screens that are loaded by the user are transmitted on an as-needed basis to the customer.
Typically this model requires the IT administrator to set aside a certain amount of bandwidth for each user connection. At the same time, administrators must be cognizant of the latency of any WAN connections that users are working over. While the protocols used to send this virtualized data to the user are optimized for the WAN, increases in latency result in increasingly large slowdowns in the users’ perceived performance of the application.
Introducing WDS appliances to this architecture looks much like any other environment – a WDS appliance is deployed to the branch office as well as the data center.
Riverbed® Steelhead® appliances can typically remove 60 – 95% of the non-virtualized desktop data from a WAN bandwidth connection. By doing so, their first benefit is to free up significant amounts of bandwidth that will allow additional users to productively use virtualized desktops. In addition, advanced WDS technologies with features such as built-in quality-of-service (QoS) that can prioritize latency-sensitive traffic, and ‘fill-the-pipe’ solutions can further improve the user experience of virtualized desktops.
The data in chart 1 shows how this is possible. Using Riverbed’s unique MX-TCP functionality, combined with its Hierarchical Fair Service Curve (HSFC) QoS technology, users are simultaneously guaranteed bandwidth for their virtual desktop connections and some of the latency impact is mitigated. The results below show operational improvements for particularly common user actions; results will certainly vary based on use case and network design.
|
