Chris Wolf

A member of the Gartner Blog Network

Chris Wolf
Research VP
6 years at Gartner
19 years IT industry

Chris Wolf is a Research Vice President for the Gartner for Technical Professionals research team. He covers server and client virtualization and private cloud computing. Read Full Bio

Coverage Areas:

PCoIP vs. ICA/HDX: A Simple Experiment

by Chris Wolf  |  October 21, 2010  |  10 Comments

Last week I traveled through various parts of Europe, while stopping for two days in Copenhagen for VMworld Europe. With eight days of travel, I thought it would be a good opportunity to test the feasibility of working from a server-hosted virtual desktop while traveling abroad. I run both Citrix XenDesktop 4.0 SP1 and VMware View 4.5 in my office lab in New Jersey. Accessing virtual desktops over the local area network using my Wyse Zero clients offers nothing short of an outstanding experience. I connect to my XenDesktop systems using the Wyse Xenith and connect to View using the Wyse P20. I’ll offer more on my experiences using zero clients in a future post, but for now I’ll say that user experience is easily on-par with a physical desktop.

Now onto the experiment. I wanted to see how both the ICA/HDX (XenDesktop) and PCoIP (View) protocols stood up to varying real-world network conditions. While traveling, I connect to my lab using a L2TP/IPsec VPN through a Vyatta VM appliance. A couple of years ago, I replaced a failed Cisco Pix firewall with the Vyatta appliance and have used it ever since.

The test itself is relatively simple. I connect to the XenDesktop and View environments one-at-a-time and perform a few basic tasks. After connecting, I open a Word document and add a comment. After that, I open a browser-based Flash application (imaginationcubed), watch the sample drawing, and then use the app to write the word “hello.” Each test lasts from just over one minute to three minutes, depending on the speed of the network I was on at the time. Also, I used Fraps to record my experience.

Note that the purpose of my experiment was not to conduct enterprise-scale benchmarking. Instead, I simply wanted to experience VMware View and Citrix XenDesktop while connecting from a variety of locations. My experiences are documented in the following posts:

To summarize, I found that as latency increased beyond 200 ms, ICA/HDX had a decided advantage over PCoIP. Text entry and menu response with ICA/HDX was still good enough to keep me productive, while the PCoIP response hampered productivity. When latency was relatively low (e.g., under 120 ms) both protocols performed well enough to meet my expectations. At several locations (i.e., Newark Airport, Munich, and Copenhagen) both protocols were able to deliver a good user experience.

When traveling abroad, there were numerous instances where I didn’t have Internet connectivity. For example, Internet connectivity was down at my hotel in Copenhagen, leaving me without Internet access for two nights. Also, I didn’t have connectivity while at several airports; connectivity was available, but I chose not to pay for it. That experience underscored the need for me to be able to run my critical Windows applications client-side while traveling abroad. While many enterprises get by with physical laptops, I have talked to others who have used client side virtualization technologies from vendors such as RingCube and MokaFive to meet the needs of the mobile user. Of course, we now have tools like Citrix XenClient and VMware View “local mode” desktops at our disposal as well. Today, the server-hosted virtual desktop often is not enough for the traveling professional (depending on access and latency considerations). It has its use, but so does complementary technologies such as SaaS applications, client-hosted virtual desktop solutions, application streaming, and of course, the trusty physical laptop. As always, the users’ requirements should dictate the technology choice.

Clients I speak with consistently state that they desire to move away from legacy application delivery models that resulted in tethering a user to a single physical device. Server-hosted virtual desktop solutions such as Citrix XenDesktop and VMware View have their place, but my experiences over the past week have proven to me that they are not the silver bullet either, and I think the vendors would agree. Client endpoint-side caching solutions such as VMware View Local Mode, Citrix XenClient, the MokaFive Suite, and RingCube also have their place. Not all client-hosted virtual desktop solutions are ready for large-scale enterprise deployments, but they are steadily maturing and will get there in the next couple of years.

What is your experience using these protocols in a variety of scenarios? I kept my tests relatively simple because I had limited time to do them. For example, I didn’t test audio/video response. We are planning to publish more detailed research that compares remote display protocols in the coming months, and I would love to hear about your experiences.

10 Comments »

Category: Client Virtualization     Tags: , ,

10 responses so far ↓

  • 1 Daniel Feller   October 21, 2010 at 8:35 am

    Interesting post and that is the best way to see if it will or will not work for you. Try it in your own and see for yourself is my advice. Try it in many different real-world scenarios.

    With regards to your mobility requirements, I have the same challenges, which is why using a hosted solution is not feasible. On the other hand, using streamed applications is much more usable for me due to their offline capabilities.

  • 2 Matt Liebowitz   October 21, 2010 at 9:05 am

    Great post Chris. Beyond the latency and performance testing, it really does show that this technology is still very much new and needs work to meet the needs of all users.

    What I really want is the Outlook in cached mode combined with Outlook Anywhere experience for my server hosted virtual desktop. Outlook in cached mode stores your email on your endpoint rather than forcing you to always be connected to the server. And Outlook Anywhere (formally RPC over HTTPS) lets you use Outlook from any Internet connection without having to worry about a VPN.

    The Outlook configuration I described above is a true “grab and go” solution. That is, I don’t need to wait for anything to synchronize to my laptop before heading out the door and running for a train. Any changes I make to Outlook while offline (deletes, new emails) are synced to the server when I reconnect to the network.

    That’s what I’d love to have for my SHVD. Something that runs on the server but is constantly synchronizing changes to my local endpoint. I can just close the laptop, head for a train, and open the VM running locally on my laptop. When I reconnect to the network, changes synchronize and I resume running off the SHVD again.

    Some pieces of this solution exist today but it isn’t fully there yet. I believe that as SHVD matures and as client hypervisors also mature we’ll get to the “grab and go” model. I’m looking forward to it…

    Matt

  • 3 Doug Dooley   October 21, 2010 at 1:45 pm

    Chris -

    This type of post is the reason why you’re one of the best analyst in all of virtualization. You seamlessly combine your research and analysis with real-world testing and hands-on experience.

    I agree with you that VDI (SHVD) pioneered by VMware and Citrix is here to stay. However, anyone who thinks it’s going to be the common virtual desktop approach for the mobile professionals is not actually using the products themselves like you are.

    Thank you for giving startups like RingCube and MokaFive the recognition they deserve. These 2 companies have very innovative and practical solutions that get better upon each release but not a lot of people know about them yet.

    Great work on this as well: http://www.DesktopVirtualizationOptimized.com

    Doug

  • 4 Chris Wolf   October 21, 2010 at 5:03 pm

    @Daniel – thanks for the feedback. Totally agree on streamed apps. I also think that’s why a lot of folks are excited about client-hosted virtual desktops from VMware and Citrix, and the complementary offerings from RingCube, MokaFive, Wanova, Unidesk, etc.

    @Matt – great point on Outlook cached mode. I think we’re heading in a similar direction with many apps. Outside of SaaS, we still need help supporting legacy apps and many vendors i speak with are trying to offer that level of application access transparency. It will be fun to watch the technology evolve.

    @Doug – thanks for the compliments. Definitely appreciated. You guys are doing a nice job getting the word out and the recognition you are receiving is well-deserved.

  • 5 Jeremy Saunders   October 22, 2010 at 4:23 am

    Nice basic test Chris, and a great post. Do you think the overheads of the VPN tunnel injected enough overhead to have an effect on the user experience? Obviously we will have to wait a bit longer for PCoIP to be supported via the Security Server, but it would be interesting to see if the ICA/HDX performance would be even better using an ICA-Proxy (NetScaler or CSG). I know that it wasn’t the point of your test, and would have been unfair for PCoIP, but interesting just the same.

    Cheers,
    Jeremy.

  • 6 Greg   October 22, 2010 at 5:54 am

    Hi,

    How could you ignore Ericom Blaze ?

    Thanks et good work.

  • 7 Steve Greenberg   October 22, 2010 at 12:11 pm

    Great practical test Chris, thanks!

    I was a bit surprised that you were good up to 200ms latency, similar tests I have done showed it breaking down sooner around 100ms, this is good info!

  • 8 Chris Wolf   October 22, 2010 at 12:30 pm

    @Steve – Actually, I used 200 ms as a generalization. That was the first test in which I saw substantial differences between PCoIP and ICA/HDX. Note that I had no tests between 122 ms and 200 ms. I was just recording the latency I had at the time I connected. Performance was acceptable at 122 ms for both protocols. Note that at 122 ms, I also had over 1 Mb of bandwidth. Also, I mentioned that my test was very simple and did not include audio/video capabilities. Your team at Thin Client Computing has done far more extensive benchmarking. I know that Shawn Bass has too. I think that the more we (as an industry) see in terms of published benchmarks in real-world scenarios, the easier it will be to make product decisions. Hopefully other folks will continue to share their experiences too.

  • 9 Chris Wolf   October 22, 2010 at 12:36 pm

    @ Jeremy – I completely agree on proxies and technologies like Branch Repeater. They are certainly a necessary part of the benchmarking process. I recommend that customers evaluate WAN acceleration technologies as part of the pilot process to see what is the best fit for their branch office environments. Early benchmarks published by Citrix have shown substantial performance improvements.

    @Greg – good question on RDP acceleration. I didn’t look at RDP primiarly because of time constraints. I was running these tests with small windows of oppurtunity in airports and hotels. I also would have liked to have tested Quest Software’s EOP Xtream under similar conditions. I travel quite a bit, so I expect to have many more opportunities to run similar tests against other protocols.

  • 10 Dan   October 24, 2010 at 2:36 am

    @Chris
    Ericom has recently introduced Blaze integration with VMware View. Ericom Blaze can now function as a VMware View client. This integration was specifically designed to address problems such as the ones you encountered on high-latency networks.

    Disclaimer: I work at Ericom