WAN Optimization and the VDI Challenge
I was sitting in on a peer-to-peer exchange about virtual desktop infrastructure (VDI) hosted by Wikibon and listening to how great VDI is for organizations. I asked about multisite VDI and what recommendations people had for implementing VDI between sites. The speaker paused and said that she hadn't see any multisite implementations. No surprise there. VDI itself is still in its infancy, but it hit me once again that if organizations are going to see VDI's benefits at the remote office, they've
February 14, 2011
I was sitting in on a peer-to-peer exchange about virtual desktop infrastructure (VDI) hosted by Wikibon and listening to how great VDI is for organizations. I asked about multisite VDI and what recommendations people had for implementing VDI between sites. The speaker paused and said that she hadn't see any multisite implementations. No surprise there. VDI itself is still in its infancy, but it hit me once again that if organizations are going to see VDI's benefits at the remote office, they've got to pay close attention to their WAN acceleration layer.
It doesn't take a genius to see why multisite VDI might sound like a really bad idea--VDI can be a bear on the WAN. Getting exact numbers is going to be tricky, but Daniel Feller, who runs the Virtualfeller blog, does a pretty good job explaining what it's like to run Xen Desktop across the WAN. He estimates, for example, that you'll need up to 85Kbits per second for Internet browsing per user and 553 to 593 Kbits per second for remote printing per user. Multiply those numbers by the duration of activity, and an estimated traffic profile can be obtained for each user. Keep in mind that those numbers can drop significantly based on a number of factors, including whether you can leverage Xen's Branch Repeater.
Lest you think Xen is alone, though, take a look at VMWare's expectations for VDI requirements. A good rule of thumb when running PCoIP is three users per 1Mb. This allows for variance in the display activity between multiple users and provides a range of bandwidth most likely to provide acceptable performance for user.
While deduplication technologies provided in WAN accelerators can address some of the bandwidth issues, it's not just bandwidth that matters here. Out-of-order packets cause retransmissions higher up in the stack that decrease application performance. Addressing packet ordering is particularly important in an application like VDI, where responsiveness will be very noticeable.
How packet ordering is treated can make a significant difference in the performance of VDI. This is a big area for folks at Silver Peak who use Forward Error Correction (FEC) and Packet Order Correction (POC) technologies to address the problem. FEC is used to rebuild dropped packets on the far end of a WAN link; POC is used to re-sequence packets that are delivered out of order.Moving up the stack, organizations also need to consider the type of VDI implementation being optimized. TCP-based VDI protocols, such as Citrix ICA, can be accelerated by a number of vendors, including Blue Coat, Riverbed and Silver Peak, but not all VDI implementations rely on TCP. Part of VMWare's PCoIP, for example, runs on User Datagram Protocol (UDP), as does Sun Ray ALP. The only WAN optimizer I've seen claim to accelerate IP and, by extension, UDP is Silver Peak.
You've also got the matter of packet flows to consider here. With a VDI implementation, every client must maintain at least one flow, and the average thin client can have 10 to 15 TCP flows at any one time. Organizations need to be sure that the data center appliance they're purchasing can support a huge volume of concurrent flow or risk scaling issues.
WAN optimizers will be challenged to identify the right application at the right time. Application classification is essential if WAN optimizers are going to know how to accelerate certain kinds of traffic. Yet many WAN optimizers continue to classify applications based on port numbers. I know this will come as a shock to those of you living in the 21st century, but classifying applications strictly based on port numbers is going to be problem when a bazillion different Web applications share port 80.
Instead, WAN optimizers need some kind of deep packet inspection (DPI) engine to look deeper into the packet to know what types of applications are being accelerated. Riverbed, for example, has long been able to differentiate between types of traffic and then prioritize user activities accordingly within the Citrix ICA environment, but that hasn't been the case with PCoIP. Riverbed claims to have addressed the issue with its AppFlow Classification Engine, which will be available in the first quarter of this year.
VDI offers organizations a number of operational benefits, and WAN optimizers can in theory help realize those benefits at the branch office. But this story will play out well for IT pros only when the WAN optimizers address all of the VDI's networking challenges.
About the Author
You May Also Like