Like any form of resource consolidation, virtual desktop infrastructure (VDI) has its tradeoffs. Enterprises must weigh the benefits of centralized administration against latency and other issues of virtual desktop performance, especially for deployments of VDI over the WAN. While users may tolerate a few extra seconds to load a file, enduring the same delay for basic desktop functions, such as typing or resizing windows, will be unbearable...
for users whose virtual desktop sessions cross continents and oceans.
Wide area network (WAN) managers who support VDI to remote locations will almost inevitably face latency challenges, said Eugene Alfaro manager of global IT operations and support at Simpson Manufacturing Company, Inc., a global building products manufacturer based in Pleasanton, Calif. Alfaro has been managing VDI performance issues with a global Citrix deployment for several years. He said many enterprises can be caught unprepared for issues with VDI over the WAN.
"You get a call from users saying [VDI performance] is slow [there]. What do they do? This is where IT will be blind," he said.
Latency is the biggest concern among networking pros considering a VDI deployment, according to an informal survey of 1,197 VMworld 2010 attendees conducted by storage vendor Xiotech and WAN optimization vendor Silver Peak. The vendors say 62% of respondents named latency as their top VDI network consideration.
A minority named other WAN-related issues as concerns, such as the ability to shape or prioritize traffic (7%) and bandwidth (6%).
More than half (55%) of respondents said they would deploy VDI outside of a local area network (LAN) environment—22% over private WAN, 27% over virtual private networks (VPNs) and 6% over a third-party or cloud computing infrastructure.
"If you're showing keystrokes across the WAN and constantly have to go back and forth to show the data, that's very sensitive to latency," said Jeff Aaron, vice president of marketing at Silver Peak. "[A VDI platform's] native compression helps and goes a long way, but that only … attempts to solve the bandwidth [issue]. Compression doesn't have anything to do with latency performance."
WAN optimization vendors such as Silver Peak, Riverbed Technology and Expand Networks claim the ability to improve VDI performance through advanced compression, deduplication, correcting packet loss and reducing protocol chattiness. Startup Wanova uses endpoint software to optimize performance by continuously syncing images.
Meanwhile, virtual desktop vendors also offer their own acceleration engines. Citrix demonstrated HDX Nitro—a collection of performance enhancement technologies, including Project Mercury, aimed at VDI over the WAN—at its annual Citrix Synergy conference earlier this year. In VMware's latest release of its VDI product, View 4.5, the virtualization giant announced enhancements to its proprietary protocol, PCoIP, to improve VDI over WAN connections.
Three ways to combat latency, improve VDI performance
When Simpson Manufacturing globally deployed Citrix Systems' VDI over the WAN in 2003, VDI performance was hardly a concern because nearly all offices were within the United States, Alfaro said. But the company rapidly grew—tripling in size between 2004 and 2008—and the WAN now supports 3,000 users in 40 locations across 14 countries.
If you go to VDI, you have to have even better monitoring and more intelligent monitoring. You don't just monitor up/down anymore. You monitor left/right. You monitor slow and fast.
Manager of Global IT Operations and Support, Simpson Manufacturing Company, Inc.
Last year, Simpson Manufacturing built a plant in Zhangjiagang, China. Rather than spend between $120,000 and $200,000 on building a local data center for the plant, the company decided to deploy VDI over the WAN, Alfaro said. But the average latency of 220 milliseconds between Zhangjiagang and Pleasanton presented a minor VDI performance problem.
Alfaro said latency of less than 150 milliseconds is ideal for optimal VDI performance, but anything under 250 milliseconds is "totally doable" with traffic shaping and WAN optimization to mitigate windows of peak congestion, Alfaro said. However, a few dozen virtual desktops at a manufacturing plant in Killorglin, Ireland, a remote town of about 1,500 people, presented a serious challenge for VDI performance. Although users are served by a data center in Denmark, the limited communications infrastructure across rural Ireland creates an average latency of 350 milliseconds and it can spike to 800 milliseconds for a peak congestion, Alfaro said.
Those latency spikes only last a few minutes per day and wouldn't be so disruptive for a single application. Users could work on another task while they wait for it to pass. But virtual desktops present the entire platform, meaning that users can't multitask during blocks of extreme congestion, Alfaro said.
"In the past, that blip of latency didn't matter and [IT managers] are going to think it still doesn't matter because legacy knowledge always gets applied today, and that's the problem," Alfaro said. "You're inheriting and bringing that into VDI and you cannot. You've got to think radically different."
To minimize latency and improve the user experience, Alfaro said his best weapons are WAN optimization, intelligent monitoring and quality of service (QoS)—in that order.
The caching function of Riverbed's Steelhead WAN optimization device has helped reduce the number of latency peaks and shorten them, he said. Citrix's EdgeSight monitoring platform has been "the best tool I've ever used to monitor the user experience," Alfaro said. The software monitors end-to-end latency and alerts him if it crosses the 250 millisecond threshold.
"If you go to VDI, you have to have even better monitoring and more intelligent monitoring. You don't just monitor up/down anymore," he said. "You monitor left/right. You monitor slow and fast," Alfaro said.
Finally, he works with his WAN provider Sprint to prioritize VDI traffic or "interactive traffic" above other applications, including voice and video. The move may seem "radical," he acknowledged, but Alfaro said users are far more tolerant of voice jitter than virtual desktop jitter.
"Because of cell phone technology, our expectation of voice is actually lowered," Alfaro said. "Yes, we have a little blip here and there but no one complains about voice, and we have voice over IP across the globe."
VDI performance complaints "will be never-ending" and not all latency problems can be solved, Alfaro acknowledged. That means WAN managers must sell users on VDI's benefits—the ability to travel without a computer, to see the same desktop from anywhere in the world, the ability to restore accidentally deleted files, the ability to quickly and cheaply standardize desktop environments, and the budget relief from having no local IT presence.
"Because you have to accept some of the performance degradation and issues that come with VDI, you also have to raise to the surface the value and the benefits to users," Alfaro said. "You've got to sell those benefits to senior management or they're not going to see [beyond the limitations]."
Let us know what you think about the story; email: Jessica Scarpati, News Writer.
Dig deeper on Application performance on the WAN