Savvy organizations are clamoring for easier access to desktops and applications from new mobile and wireless clients that, despite their small size, put big demands on the data center. The push for mobility to go beyond laptops to handheld devices means that centralizing desktops and apps to run from the network is more than a good idea -- it’s a new requirement.
But burning enthusiasm for virtual desktop infrastructure (VDI) can become a headache if the data center isn’t built to meet users’ expectations. What has worked just fine for hosting virtual servers won’t translate to hosting virtual desktops or applications. Unique demands on storage and networks mean retooling some key components. When designing your data center for optimal VDI performance, make these VDI solutions your top priorities, test them thoroughly and begin working toward specific and measurable performance results.
Focus on user experience
Even fast laptops take too long to boot up; most of us can make a round trip to the coffeemaker before a login screen appears. Over time, apps and utilities layer on top of each other and slowly bring things to a crawl, a phenomenon sometimes called “Windows rot.” Virtual machines (VMs) can beat dismal startup times, as long as the back-end design is ready to deliver.
Tweak storage performance
Drive thrashing, where a PC drive light burns furiously at startup, is a necessary evil for PCs, laptops and servers. VDI files have a shared design, meaning minimal files are needed to boot up. Intensive drive access cuts down on drive thrashing for VMs but is a real threat to performance when multiple users log in simultaneously. To better prepare for peak loads, one of the VDI solutions entails boosting the I/O throughput of the storage area network (SAN) by using upgraded data switches along with multipath, load-balanced connections. This applies to both Fibre Channel and iSCSI/GigE SAN, and it eliminates a choke point during heavy loads.
Separate and layer disk storage
Virtual desktops use a “golden image” template at startup. The template is a large file but doesn’t require a lot of storage, as it’s the single shared master copy. It still, however, requires fast drives and I/O to stand up quickly and therefore belongs on high-end RAID arrays. Conversely, user data should live on cheaper storage -- big, slower drives that are simple to install and manage, such as Dell EqualLogic. Separating tiered storage using these drives for transient user data allows the critical system files to run at optimum speed while eliminating some hardware costs. An added bonus is that when the golden image is refreshed, user data remains untouched.
Plan for real-world testing
Vendor white papers and marketing slide decks promise high ratios of virtual desktops on each host server, but no one knows what an old, fat client loaded on dozens of Windows VMs will do to a server’s memory and CPU. That has to be tested in the real world. Instead of using formulas for planning server builds, load up legacy apps, clunky printer drivers and spaghetti-coded scripts in the lab and capture the results before committing to a production design. You should, of course, be ready to quickly scale up when an unexpected event occurs or when the demand for virtual desktops increases.
Invest in app performance management tools
In the past, WAN speed was measured from point to point. Today, it means measuring through uncharted and opaque cloud computing services or across shared, multivendor WAN connections. My company’s many years of working on local LAN desktops set high expectations for performance and fast response to service interruptions. Mobile computing response times will benefit greatly from VDI, and that is a critical selling point. But getting comparable performance from uncontrolled access points worldwide through VPN and wireless broadband clients requires explicit and targeted monitoring.
Measuring end-to-end performance (from the server to the user and back) requires a new suite of VDI tools. Because of the many hops from host to client, you must break out the layers and segments of the network/WAN connection and analyze each. By using a product like Firescope monitoring suite, you can pin down the root cause of glitches or outages, ideally before users feel the pain.
Consider pooling VMs
Pooling desktop VMs is one of the elegant VDI solutions for scaling up or down as needed to manage capacity. Instead of running virtual servers 24/7, a user login will call up a VM desktop or app to start on demand, and then stop again on logout releasing the license and resources back to the server. Pooling can be seen as a new kind of concurrent system that can reduce power, cooling, licensing and overall need for infrastructure. Creative types are rolling out some clever technologies that let user configurations follow this new login and logout behavior, like Appsense, which avoids cumbersome roaming profiles.
Here is a list of particularly useful resources if you're considering a shift to VDI:
Using VMware RAWC
What did you think of this feature? Write to SearchDataCenter.com's Matt Stansberry about your data center concerns at email@example.com.