This blog is to respond to questions from my Citrix, Microsoft, HP: Best Practices for Scaling Virtual Desktops webinar on June 17, 2010. I received 22 questions during the webinar and unfortunately did not have time to answer them all. I will cover the first half of them in this blog and address the remainder in a follow-up blog. Some questions with similar content have been combined into a single question.
Q: Do you have a link to the specific entry on your blog about your experience with the issues between NIC teaming drivers and the Hyper-V role?
A: Yes, though technically it is not on my blog, but on my XenDesktop with Microsoft Technologies website. Under the Other Resources section you will find the entry “How-to Guide for HP NIC Teaming Software and Microsoft Hyper-V”. Just follow that link.
Q: What would you estimate the ratio of virtual machine to server CPU core to be?
A: It depends mostly on the processor. My experience with Intel Nehalem processors is that the maximum VM Density per core is around 11 for a medium workload. However, I would not recommend running at 11 for that workload, I would probably plan for 9-10 per core to leave some headroom for burst capacity. Keep in mind also that those numbers are taken without an anti-virus solution in place.
Q: Do you have best practice hardware and software setups for 50-100 user increments that are scalable for adding additional 50-100 user blocks that can be downloaded?
A: Not at this time. Citrix Consulting Solutions is currently working on a concept document that covers the creation of a small scale architecture that can be repeated. That document is in the final stages of review, so it should be released in the near future.
Q: From what I know we can create two different types of profiles – standard (for everyone) and dedicated. The dedicated one gives a users “administrator” privileges to make permanent changes to their desktop that are persistent. For a “standard” image, how much information follows a user on their profile from session to session as they log on and off multiple times?
A: Good question. What you are actually referring to is what we call vDisks for Provisioning Services. The vDisks are “golden master” images that are shared or private to a virtual machine. When a vDisk is in “private” image mode, all changes are permanent to the VHD. The “standard” image is considered “read-only” and all software installed is lost when the virtual machine is rebooted. The standard mode is used for creating pools of desktops that are available to service any user. The amount and type of information that follows a user between sessions is dependent on how profiles are managed. The key to deploying this successfully is using profile management software that can redirect the user data folders to a network location or save off the data in a roaming-profile type scenario. In both image modes, the user can be configured to be either a local administrator or just a local user, similar to any other workstation on the network.
Q: How different would the architecture be in a combined scenario of XenDesktop and XenApp?
A: That is the beauty of using XenDesktop Enterprise or Platinum editions, they come with XenApp and the only difference is the addition of XenApp servers. The XenApp farm can leverage the existing License Server, database server for the hosting the farm (though it will be a different farm), and even the same Web Interface servers. In fact, most XenDesktop deployments today include XenApp. Using XenApp provides the advantage of offloading the processing for resource-intensive applications and also helps reduce the licensing costs of applications that are not necessary for all users to have installed.
Q: How many virtual desktops could be supported with a DL380 G6 32GB RAM and 8 cores?
A: Let start by saying “it depends” probably more on the user workload and the SAN/Local storage available than on the RAM and Processor power. I cannot tell you how many desktops you could get in your environment, but I can tell you what I have seen with regard to limitations. The processors will not be a limiting factor because Windows 7 core density on a G6 (Nehalem class processor) will be around 10:1, so 8 cores would about 80 machines. Your primary limiting factor appears to be the RAM. With only 32GB RAM you could get a maximum of about 30 Windows 7 desktops (1 GB RAM) or maybe 60 Windows XP (512MB) on Hyper-V. A DL380 G6 supports up to 8 SFF drives, so the local storage could be used to host guest virtual hard disks. Once again though, it all depends on your user workload.
Q: Was that Hyper-V on server core?
A: No. We used the full Windows Server 2008 R2 Hyper-V version instead of server core. The primary driver for this decision was the need to obtain detailed performance metrics on the host servers and correlate that data with our internally developed test harness, called STAT. Unfortunately, the STAT agent responsible for the metric collection requires a GUI install which would not work on server core. A nice benefit of not using server core was of course that I didn’t need to figure out the script commands for installing and configuring HP NIC teaming drivers.
Q: Can a presentation be prepared or at least a comparison between Hyper-V and ESX4i (VSphere)?
A: Yes. I am currently planning a similar topic for my Q3 webinar. The VMWare EULA prevents any direct comparisons or performance results from being communicated, but I plan to cover the best practices for running guests on the three hypervisors supported by XenDesktop: XenServer, Hyper-V and VMWare VSphere.
Q: Which FlexCast model applies to your testing and scalability numbers?
A: The model that I used is considered the “Hosted VDI” model where a full VM is running on a Hyper-V host. However, technically speaking, it includes the “Streamed VHD” model, since the VHD for the pooled desktops was getting streamed to the virtual desktops running on the host system.
Q: The response times are based on local LAN-based users, correct?
A: Yes. However, since the response times are calculated by Login VSI from within the XenDesktop session, they would be the same no matter where the desktop is accessed from. For WAN users you would need to adjust for the latency on the line.
Q: Seems like Citrix talks a lot about deploying on Hyper-V. Is there a reason why this is done on Hyper-V instead of using XenServer (or even VMWare)?
A: Good Question. The answer is that we perform these tests on all hypervisors. It just turns out that I am the XenDesktop on Microsoft architect, so I volunteered to be the one that did the Hyper-V testing. We do have some scalability test results for other hypervisors on our Citrix eDocs website. Once there choose XenDesktop >> XenDesktop 4 >> XenDesktop Scalability Guidelines. We are continually testing XenDesktop with all supported hypervisors and normally make that information public as it becomes available.
That concludes the first round of unanswered questions from my Scaling XenDesktop webinar. The rest should be posted within the next few days. I hope you find the answers helpful. If I misunderstood your question, feel free to clarify it as a comment below.