back to article Sizing a server to support desktop virtualisation

Virtualising the desktop can bring benefits at the endpoint. It makes desktops more manageable, can reduce power load throughout the building, and can make systems more secure. But IT departments shouldn't underestimate the additional investments required at the back end. Running more endpoint logic centrally can have huge …

COMMENTS

This topic is closed for new posts.
  1. Anonymous Coward
    Anonymous Coward

    net booting browser

    what's wrong with a network booting linux box with just a local browser and a remote mounted profile / home directory.

    That's what we are in the process of doing at least - all the apps for for these users are through the web browser anyway.

    Upgrades are just a quick reboot away, management is centralised, server overheads are low.

    1. SilentLennie

      webtechnologies and servers

      I've been arguing that for years, it is much easier to manage a few webapplications on a few central servers than having to deal with all those desktops.

      But I do not believe in cloud-computing very much, I think running your applications on your own hardware makes a lot more sense.

      If you have many locations of the company, running your own hardware centrally or possible at a datacentre is a possibility. But I prefer to not have others deal with our data.

  2. Tezfair

    Doing this now

    I have a 2 year old server that I setup as 4Gb on a 64 bit server 2003 dual core (no cpu virt) to have a RD / VPC play around whilst accessing it remotely. As soon as 4 desktops were running the server the lag was horrendous.

    Although the system wasn't committing all the ram and the HD activity was negligible the cpu was 100%

    I have to quote a company a RD server that can handle 20 desktops for branch office access. The server requirements simple means its cheaper to buy separate PCs.

    The licensing for something like xenserver is 4 digits

    I simply don't see much call for Virtualising the desktops, after all, you then have to pay twice for the OS, one on the physical machine and one on the virtual.

    1. Anonymous Coward
      Coat

      I've rambled a bit. Sorry.

      Tezfair, I humbly apologise if you're aware of all of these things, but your post sort of suggests that you aren't.

      Paravirtualisation shouldn't really be compared to hardware enabled virtualisation. There is a definite performance difference. Doubly so on modern hardware. Typically we find at work that it's actually the number of spinning disks, rather than CPU constraining an average virtual server host box - provided you put in a decent Xeon, or pair of Xeons.

      If you're stuck with Windows, for your client, you may also want to look at Microsoft's RDS (Remote Desktop Services, previously Terminal Services). The licensing is a bit easier to swallow than VDI. All you need is Windows Server, the relevant RDS CALs, and you can re-utilize even the oldest hardware - provided you can run a RDP client - which there are for most platforms now. You could even consider using thin clients, if new client hardware is required. Generally speaking it normally works out cheaper than VDI for smaller numbers.

      The problem is the nature that it's a shared box, and it may be a hurdle to get over for users, that they shouldn't be able to do what they want with their session. However, if you're from a *ix background you'll probably appreciate this and can deal with the situation and the users.

      On the subject of the cost of the OS - again, Microsoft Windows have very specific amendments for virtualisation in the recent server and desktops licenses. For example Windows server 2008 standard allows you to run the same license on the host and in a single virtual machine. Enterprise upto 4 virtual machines, and datacentre is unlimited. There is also Windows Hyper-V server - a free, "GUI-less" (I use that term loosely) edition.

      If you're not using Windows for the host box then the licensing applies as necessary.. Last time I played with it, Windows Server ran acceptably under Linux's KVM implementation for light-medium weight loads. VMware, yeah you're right on that front for the cost :)

      If you're going for VDI then the licensing requirements are a bit more complex. I believe it's a requirement for volume licensing - although it's been a while since I've had to do a VDI install, so I may be a little out of date and off on that.

      I'll get me coat.

  3. The Cube
    Stop

    Desktop virtualisation is a solution in search of a problem

    Lots of people want you to buy desktop virtualisation, conspicuously these are the same people whose gravy train from you implementing server virtualisation is now drying up.

    Tin vendors want you to buy lots of new servers, network and storage to support it, and that is all nice high margin stuff too not like the utterly commoditised desktops.

    Virtualisation software vendors are all wondering why people aren't buying the new way to spend money on their technology and stuffing it with more features as if that addresses the underlying lack of real benefits or demand.

    Desktop virtualisation is being dressed up as solving any problem you can think of from "support cost" to democracy in Libya if the vendors think that is what will persuade customers onto the next gravy train of "migration" or "refresh" spending. Well, guess what, management tools and deployment technologies are advanced enough that it makes no real difference where the disk that the user's machine is written to lives from a management time, complexity and cost perspective.

    What does make a difference is the power bill, a well configured modern desktop or laptop will use a lot less power than the 100W ++ per machine used to try and make the numbers work for desktop virtualisation, a good desktop unit now uses less power than the monitor you are going to plug the thin client into anyway. Once you have saved the 25W per desk on the desktop (9 to 5 only) you then have to power not just more servers but more network kit and more storage kit in the data centre to serve those virtual desktops. Even worse, most of that kit will be on 24/7 because if any user is working most of the virtualised desktop platform needs to be running, even when it does manage to go to sleep the overheads of the data centre are still there costing you money.

    There are some corner cases where this makes sense, there are always some apps better delivered off a server which have not yet been migrated to run in a browser. There are always some environments where some other reason such as physical security dictates the removal of the desktop machine.

    To people hawking desktop virtualisation as a mass market "solution" to a problem: This horse is both dead and tired of being flogged, you've made your money selling server hypervisors where they actually made sense, now find something else to sell, dressing up desktop virtualisation to pretend that it is a majority solution is just lipstick on a pig.

    1. Anonymous Coward
      Anonymous Coward

      VDI

      I don't think it is a solution in search of a problem at all. Say, for example, you have a set of users that need to all run identically configured machines for an intensive calculation process. Some of the code is 3rd party, some of it in-house. These machines need to be high-spec and high availability. You then need "machine" isolation because you're using parallel computation in Matlab and Matlab can grind the shit out of whatever you put in front of it.

      High spec local machine = fails on high availability, remote access etc and constant config.

      Terminal Services = non-starter.

      VMs = possible but updates then become an issue with more machines.

      VDI using single base image + user session delta storage = Just what the doctor ordered. You get the isolation of a VM with the single base that's easy to update. High availability, constant config, scaleable through create on demand etc.

      I'm willing to be educated on this but I think there are definitely valid use cases and I have one. It is by no means a solution for everyone but don't go thinking it's a solution for no-one.

This topic is closed for new posts.

Other stories you might like