For the past seven months, my team has been evaluating VDI offerings in hopes of achieving the management benefits promised by desktop virtualization vendors. During that time, I learned that even the latest and greatest products from top desktop virtualization vendors have limitations that some organizations may not be able to overcome.
The company I work for has been a VMware customer for several years, and we have successfully virtualized about 95% of our data center server infrastructure. Since VMware is established at my organization, VMware's View product was a logical choice.
Even though VMware offers a great server virtualization product, a virtual desktop environment is completely different than a server environment, so I needed to evaluate an alternative. Since Citrix has a proven track record with XenApp and has invested considerable resources to develop its XenDesktop offering, it was the other obvious choice.
Choosing established vendors with solid track records was an important factor because moving from physical to virtual desktops is a costly, time-consuming undertaking, and the last thing we can afford to do is choose a vendor that is not established. Of course, well-established companies do discontinue unsuccessful products, and they do fail, but I am willing to bet on the established companies with roadmaps and investments dedicated to improving their virtual desktop offerings.
VMware and Citrix both sponsored technical resources for this proof of concept (POC), and Citrix even sent technical resources on site four times to assist in our POC. I was very impressed with Citrix, especially since we did not have an existing Citrix environment and no commitment to purchase. VMware was also very accommodating, and when our engineer had questions, its representatives were quick to respond.
During the pilot, we went through two releases from both vendors and were careful to compare the latest versions of both products fairly. The final pilot involved the latest releases, VMware View 4.5 and Citrix XenDesktop 5.
The POC virtual infrastructure
A large pilot environment wasn't necessary because my goal was to find out how the systems would perform on the wide area network (WAN). Virtual desktops introduce a massive amount of traffic.
Our WAN testing consisted of 10 zero clients, five for View and five for XenDesktop. We also tested thin clients. I was more interested in zero clients, however, because those devices require less management. I began our WAN testing on our smallest, most latent lines, which were 1.5 MB, 80 milliseconds (ms) one way, and moved to our largest, least latent lines -- 10 MB, 3ms one way.
The back-end hardware we used was new and identical for both environments. We didn't introduce solid-state disk because the pilot environment had only about 60 virtual desktops. The Raid 5 array was configured independently of our other environments and included 146 GB 15,000-rpm Fibre Channel disk drives.
PCoIP, HDX and the end-user experience
During my research, I read that PC-over-IP (PCoIP) took more bandwidth than ICA (now HDX), but that wasn't the case. If we tweaked the settings to match those of ICA, it actually took less bandwidth. Not a lot less, but I was surprised.
Of course, both environments can be manipulated to get different bandwidth numbers, but even with bandwidth changes, the resulting performance was not good on the 1.5 MB lines for either product.
The Citrix environment outperformed the VMware one on those lines from an end-user perspective. Although VMware View took slightly less bandwidth, PCoIP did not perform well when scrolling. XenDesktop did a better job when scrolling through different applications and the Web. But neither could support video on the 1.5 MB lines, so we quickly shut down video testing.
The Citrix clients showed a cleaner image on the monitor, while VMware clients looked blurry because of the different protocols behind PCoIP and HDX and how they render images. In fact, a couple of our test users said they got a headache looking at the screens running View.
After three full days of testing on our 1.5 MB lines, we concluded that virtual desktop infrastructure (VDI) would work only if the location had five people or less and none of them needed to do anything that involved Flash, HTML 5 or Silverlight.
On an average day, those locations use about 20% of their allotted bandwidth (one location had 30 users, and the other had 10). After introducing VDI, we used nearly 100% of that bandwidth with five machines doing random work such as surfing the Web and working on a Word document in both environments, so we wrapped up testing.
While I was disappointed in the results from the initial tests, I hoped to get better results at the 10 MB location, and the initial results appeared promising.
But once again, scrolling in the VMware View environment was not as good as in the Citrix environment, and XenDesktop 5 was more responsive than VMware View 4.5. VMware, however, did a better job with the video content we tested.
With video testing at the 10 MB location, we did not redirect any video traffic; it was rendered on the server, and both products used large amounts of bandwidth when playing video. On average, when playing a video from CNN.com, we saw 5 MB to 7 MB used in both environments (not full-screen). When testing the same video on a thick client, average bandwidth was 900 KB.
At that point, I was convinced that a zero-client architecture was impossible if we wanted to do any video over the WAN, unless we were willing to do some major upgrades to our WAN links.
I also decided to discontinue testing View because we weren't getting the results we were looking for.
Thin clients vs. thick clients
Our testing outside of video was at least as good, if not better than, the experience on the current "thick client" architecture. Without the introduction of video, each device used an average of 400 KB while working in the VDI environment. So, we began testing thin clients and thick clients with client software to take advantage of redirect technology.
Although I was discouraged that we could not use zero clients, I was looking forward to what Citrix's HDX technology could do for video redirection. We made the necessary changes to the system to force redirection, but the performance and bandwidth usage didn't improve.
With thin clients, round-trip latency increased from 6 to 8ms to 30 to 40ms. What we didn't realize was that redirection will work only if round-trip latency is 30ms or less, and we were now exceeding that latency threshold.
We could deploy Citrix VDI to end users who don't use video, but I'm concerned that they may need to use it in the future as video training and Web conferencing use grow. So after seven months of testing, we determined that neither VMware nor Citrix VDI products could satisfy the needs of our organization.
Today, it seems the only benefit that virtual desktops offer is data centralization. But I will continue to monitor VMware and Citrix's VDI products, and when they have made enough progress, I'll revisit those offerings because I still believe the technology will be applicable in the future.
ABOUT THE AUTHOR:
Jeff Moore is an IT Manager in financial services with nine years of experience in research, development and infrastructure planning.