Consolidate for Small Scale


The Ask the Architect email inbox is getting quite large and I’ve received some great questions. I thought I would answer a question around the small-medium business (SMB) space, as it relates directly to a recently delivered Virtual Desktops for the SMB TechTalk and the reference design included within the XenDesktop Design Handbook.

Namely, the SMB white paper focuses on use cases of 200-500 desktops, but what if you only have 50-100 desktops? Do you need all of the servers? Yes in that you will need the components if you plan to do single images but no you won’t need all of the physical hardware. For example, in the reference design, there were the following infrastructure components:

  1. XenDesktop Controller/Web Interface (*2)
  2. Data Store/License Server
  3. Provisioning Server (*2)
  4. NetScaler VPX (*2)

Of course all of these are virtual servers. If we are talking about 50-100 desktops, one virtual server could easily contain items 1, 2, and 3. Ideally, you would have a second VM, on a different physical server, which would contain 1 and 3 to allow you to have some level of fault tolerance (this is a decision you have to figure out). If you don’t need intelligent load balancing, remove the NetScaler VPX. Even if you need secure remote access, you could either get an Access Gateway virtual appliance or run Secure Gateway.

Once that is done, just distribute your windows desktops across the remaining servers.

It basically comes down to this… Nothing is preventing you from putting all of these items on one virtual server. If you are sub-100 desktops, that might be the best way to go to better consolidation. If it was me, I would still follow the guidelines in the TechTalk and white paper as making these separate VMs. You still get consolidation but have greater flexibility for potential future changes and can better optimize the OS for the role it is being asked to do.

Daniel – Lead Architect

Advertisements

13 thoughts on “Consolidate for Small Scale”

  1. Hi
    First off great SMB document and techtalk! I am currently facing a customer with 150 employes wanting to go Xendesktop. I have used your SMB doc to size the servers but i wonder about your choice of Windows OS for the datastore – Wind2003SP3. Is there any reason not to go Win2008 – increased overhead etc.?

    Like

    1. No real reason. The only commonality between the solution and databases is that we need to use SQL Server 2005. If you use SP3 then you can go with W2k8. But really nothing would prevent us from going to W2k8

      Like

      1. thanks df – great stuff:)
        Also – it seems the customer is put off by the hardware cost going from a pure Xenapp environment with a published desktop. So i was considering these options:
        1) maybe consolidating the provisioning servers in the Xenserver Infrastructure Module – hmm would that work in a 150 users environment?
        2) Going with DL 385 servers with AMD Opteron Procs instead of the DL380. This is only because it allows for 24 4gig RAM modules instead of having to go for 12 8gig modules (the 385 has 24 DIMM slots as opposed to 18 on the 380). But the big question here is will the AMD Opteron processor perform as well as Intel’s Xeon 5500 series? Would you have any experience with this?

        thanks for sharing this great stuff with us:)

        Like

  2. We’re doing a mini pilot/early rollout of Xendesktop for a 50 user office, about 10-15 of them this year on a single server and then adding a second server next year for the rest and to add some resiliency. Glad to have some theoretical validation of our methodology!

    Like

  3. Moving PVS to XenServer will work in small environments. Basically, the hypervisor adds a level of overhead that limits the overall scalability of PVS, which is heavy on network. But because we are dealing with a smaller environment, it should work.

    As for the discussion between AMD and Intel. Honestly, I have no idea. I’ve read some papers that said the AMD Opteron 6100 is faster/cheaper than the Intel Xeon 5600s, but those were in instances where you were able to fully optimize the processing across all of the cores. So the question is if XenServer would show the same benefits. Unfortunately, I haven’t seen any tests nor have I seen any customers use the Opteron 6100s yet and push them hard.

    Like

    1. I think we will go with moving PVS to Xenserver then, we can always move to physical boxes in case performance sucks. Also it was suggested going to NLB clustering of the Controller/WI servers shaving of the Netscaler cost.

      I still need to go deep diving into google space for that Opteron/Intel comparison on Xenserver – but info seems hard to find.

      Thanks for a great site – has really helped a lot:)

      Like

  4. For the NLB clustering, that will work. you just want to make sure that it is monitoring appropriately so you don’t get false positive or negatives. I’ve seen some customers bit implement an intelligent load balancing solution and end up with outages.

    Like

  5. Yes i do realize that a NLB cluster could potentially do more harm than good! And i guess the question is whether failover will ever occur – chance being that a crash will occur at the application level and not the server level.

    Regarding the Netscaler VPX design – I was wondering why having a HA pair for internal WI/Controller loadbalancing AND a another pair for the external access gateway role? Why not just use a single HA Pair and route the access gateway traffic to the virtual appliance? I havent tried this but i guess using vlan trunking this should be doable and secure even though the appliance is virtualised?

    Like

  6. Unfortunately, I’ve seen too many cases where a failure in the SW or HW caused the server to fail. It is really just a decision on the risk of failure. If you had thousands and thousands of users, the risk is pretty high. If it is 100, the risk might not be as great.

    As for the VPX, you are right, you can simply use a single pair and have one of the ports for DMZ items and another port for internal and simply block it off within the NetScaler so the NetScaler is not spanning the internal FW. Most security people would see that and freak out, although when you get down to it, it is a good solution.

    Like

  7. “it seems the customer is put off by the hardware cost going from a pure Xenapp environment with a published desktop”

    The customer is correct. They should look at their users need and provide a hosted desktop or a VDI machine depending on the user scenario. It’s way cheaper to publish a desktop. Always start by having the customer defining their users, else you won’t be able to to provide them with a cost efficient platform that suits their users.

    I asume you will use XenApp in the environment anyways so it’s pretty simple for you to provide a published desktop as a second option to the VDI’s.

    Like

    1. Hi Daniel

      I saw your webcast the other day on hosting < 250 desktops on local storage and would like to go ahead with this. I note that your design called for 8 x 144GB disk in a raid 10 configuration which would give me around 550GB of usable space and your talk said to use MCS as this reduced the cost as no need for PVS.

      However, I am confused by some of the articles I read which say that if I don't use NFS then thin provisioning is not possible and so a full clone of the virtual machine is made rather than reading from the "gold" image and then using differencing disks.

      If this is the case then I dont see how I can get 50 virtual desktops on 1 physical host with the above disks using XenServer as a hypervisor. I'd like to use XenServer as a) I prefer it, b) it gives me true end to end support and c) its included inthe price for XenDesktop.

      Are the other articles I read incorrect and XenServer will use differencing disks to keep the space requirement down or is it the case, for example, that I would need to use ESXi as the hypervisor which would allow thin provisioning off local storage ?

      Any advice you can give would be greatly appreciated.

      Many thanks

      Phil

      Like

  8. Thin provisioning is just based on the underlying storage infrastructure. You can change the local XenServer storage to provide thin provisioning functionality. This post shows how: http://www.xendesktopmaster.com/how-to-enable-thin-provisioning-on-a-local-xenserver-5-5-storage-repository/

    Also, if you are using XenServer 5.6 FP1, it is quite a bit easier:
    http://community.citrix.com/display/ocb/2010/11/19/XenServer+Thin+Provisioning+on+local+disks+%285.6+vs+5.6FP1%29

    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