Hosted Shared Desktops or Hosted Virtual Desktops?


This is probably the most common question I get when going through a virtual desktop design… Should I use the hosted shared or hosted virtual desktop model? For once, this isn’t an “It Depends” answer. This is a “Yes” answer. You will mostly likely need them both. For those of you who aren’t sure what the difference is, it is pretty straightforward:

  1. Hosted Shared Desktops: A published desktop on XenApp. Users get a desktop interface, which can look like Windows 7. However, that desktop is actually being shared by every user on the server. Although we can configure restrictions and redirections to allow users to have a smaller impact on each other, there is still a risk. Many users to one desktop.
  2. Hosted Virtual Desktops: A Windows 7/XP desktop running as a virtual machine where a single user connects remotely. One user’s desktop is not impacted by another user’s desktop configurations. Think of this as one user to one desktop. There are many flavors for the hosted virtual desktop model (existing, installed, pooled, dedicated and streamed), but they are all located within the data center.

The big reason why people want to figure out if they need a hosted virtual desktop is because of scalability, which equates to money. I can get 100-200 concurrent users on a Hosted Shared Desktop model and maybe 50-100 concurrent users on a Hosted Virtual Desktop model with the same hardware. Seems like a no brainer, go with Hosted Shared Desktop.

Unfortunately, it isn’t an either/or answer. It is more of a 70% one way, 20% another way, and 10% a third way.

To do it right, you have to start by understanding your users. Citrix calls it User Segmentation Analysis, but it essentially means gathering information about my user groups to understand what they need to do their job. If you do this, you will see that the decision isn’t nearly as difficult as you expect. Here are a few examples and how I would align the groups with the most appropriate virtual desktop model (and I’m mostly looking at the application aspect, but we would also want to look at user location, mobility and end points):

Group Description Recommendation
Group 1 Users are mostly within one or two applications all day. This application is the main line of business application. Their performance is based on speed and accuracy. Hosted Shared Desktop
Group 2 Users have a core set of applications they require to do their jobs. Oftentimes, these users must be able to modify system-level settings like environment variables, or install their own applications Hosted Virtual Desktop (Dedicated)
Group 3 Users focus on content creation utilizing Microsoft Office and Adobe Photoshop. They users also browse for content and graphics online via a browser. Hosted Shared Desktop
Group 4 Users utilize a few applications that consume significant amounts of CPU resources when doing certain activities (video rendering or code compiling) Hosted Virtual Desktop (Streamed to Blade)
Group 5 Users utilize a certain application that requires admin-level privileges Hosted Virtual Desktop (Pooled)

I prefer to start with the most scalable solution first, as long as it meets the user requirements. That is the key point… user requirements. Many users need additional functionality or capabilities that are not suitable for the hosted shared desktop model. Once you reach these users, you need to figure out how to provide them with the most appropriate desktop.

In the end, there is a balancing act that goes into the design. If I have a small group of users that can utilize 3 different models, and 1 of the models is already in place, then it only makes sense to have those users use that model. It simplifies the infrastructure and makes it easier to manage.

Daniel – Lead Architect
XenDesktop Design Handbook

Advertisements

8 thoughts on “Hosted Shared Desktops or Hosted Virtual Desktops?”

  1. Dan,

    One other thing I see over and over is server side rendering of multimedia. Have a few users launch a Flash/WMV video from youtube and see it crush the processor on the server. However if the target device supports the client side rendering the server breaths much easier. Unfortunately lots of folks see the Zero clients as the ultimate device because of the lower client TCO but might fail to see the increased server costs. At least in an host VM solution, if you peg your vCPU, you’re the only person complaining.

    Like

    1. Agreed and great point. I’ve run into a few deployments with some less than stellar results when using certain thinclients because they dont’ have the power to handle the multimedia. Flash is one of the worst. It uses a lot of CPU cycles. Has an impact on the server.

      Like

      1. Can you comment on why XenApp Hosted Shared Desktops perform much better over higher latency network conditions than XenDesktop 4 pooled hosted virtual desktops?

        In my testing using XenApp 5, the end user experience is great even over a 600 ms latency connection. When I attempt the same use case with XenDesktop 4 on the same end node (Windows 7 laptop), the user experience is awful (typing lag, Windows multi-media pauses/freezes).

        I’ve heard rumors that the ICA protocol is different for XenApp vs. XenDesktop but I don’t understand it.

        Any help is greatly appreciated.

        Like

    1. Dan,

      The XenApp configuration is Citrix defaults. We changed the ICA blob for XenDesktop to prioritize the following:

      CTXCAM ,0
      CTXSBR ,1
      CTXMM ,1
      CTXFLSH,1
      CTXGUSB,1
      CTXSCRD,1
      CTXCTL ,1

      We also spent a lot of time optimizing the XenDesktop Windows XP gold image for performance by following your best practice document.

      Like

      1. Wish I could give you a good answer as to what is causing it. My suggestion is to run a perfmon on the XA and XD desktops capturing the same HDX metrics. That should at least tell us what portions of the protocol is using more bandwidth.

        Like

  2. Hi Dan, the licensing impact of HVD is quite significant for users, especially those who want to use thin client devices. Also as SPLA doesn’t include desktop licences if you want to offer DaaS you look to be tied to an RDS solution.

    How do you see DaaS providers cope with this licensing impasse, are desktops in the cloud for rent possible under Microsoft eula?

    Like

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s