Take the 2-minute tour ×
Server Fault is a question and answer site for professional system and network administrators. It's 100% free, no registration required.

Does anyone know of a good solution to have a Windows desktop standby so that users can be back up and running quickly in case of hardware failure? Assumption is that the desktops aren't generic but highly customized for individual users.

Although I'm aware of bare-metal restore products out there, I'm looking for alternatives which would allow the standby to simply be booted/brought up without any delay. A bare-metal restore solution would be fine as every day type of backup when one only expects a desktop failure once in a while (or even a few failures a day). But this is for a DR type setup, where dozens or hundreds of desktops in a different site would need to be activated at the same time.

The backup/standby desktop doesn't need to be fully up to date. Ideally, the solution would allow scheduling periodic updates.

Finally, I realize that having VM desktops is a way to do this (by replicating the back-end storage). However, I'm yet to see a mature solution that will allow for powerful quad-core VMs.

Edit #1: Sure, all the desktops start with a generic image that has things like Office, and so on. However, there are many additional applications which go to specific users only, many of them aren't generic. They also often auto-update from the vendor, meaning that even if a second desktop is built with the right applications, only if they aren't used, they will fall out of date and not work.

I've never worked in an environment where desktops are truly generic. I can think of instances where this is a reality, but come on people, this is not the answer here.

Edit #2: Lots of curiosity (below) surrounding the requirements. Agreed, they may be unreasonable, but then again, if there's a way to do this well, I would spend the money. The question really is can this be done well, and if so how? It should be obvious from the initial question that I've already thought about this quite a bit, as well as looked at possible solutions. Now, as to why would I spend the money on such a perfect solution, because this is Wall Street.

share|improve this question
add comment

10 Answers

I prefer a two-fold solution - Not "highly customizing" PCs to individual users, and having spare comptuers ready to go.

If you are working in a large enough organization to need to do DR for hundreds of PCs, I find it shocking that you're letting people use "highly customized" PCs. I'd think your desktop support costs would be eating you alive.

I roll out factory clean configurations that are modified by startup scripts upon joining the Active Directory domain, and then load software on top of them via group policy. If I need to "DR" a PC, I just put a computer on the desk with a factory clean image and join it to the domain, or re-image the existing PC with a factory new image.


If you have to do this, you might look at "Clonezilla Server Edition" (http://clonezilla.org/clonezilla-server-edition/). It's an open source solution that can PXE boot and can clone NTFS volumes / multicast transmit images, etc.

share|improve this answer
1  
Our solution to this problem is similar to this. We really only have 4 different kinds of computers in circulation, Precision 380s, Precision 340s, Optiplex 755s, and Optiplex GX400s (On their way out). If anyone's computer dies, we just swap the hard-drive out into a new one, and get the parts shipped out to us 3-5 day from Dell. Having a very small group of base-hardware to work with is extremely nice. We can install the drivers for each machine very easily, run SysPrep, get a product key on the machine from a COA, and join the domain. In the future, buy similar hardware if you can. –  phuzion Jun 18 '09 at 13:42
    
I prefer such a solution too, but it isn't applicable given the business requirements. –  Toto Jun 18 '09 at 15:24
1  
How can you be managing "hundreds" of PCs that way a reasonable total cost of ownership? Sounds like heavily flawed "business requirements" to me. You absolutely can roll out the vast majority of your software with automated installation mechanisms, including vendor updates. You're not trying hard enough-- anything can be automated. What do you do when you replace computers? Build everything by hand? Automation IS the answer, because it does the job more efficiently. –  Evan Anderson Jun 18 '09 at 15:33
    
Having said that, take a look at Clonezilla if you want something open source. That'll do what you want, re: re-imaging, though gathering the images might be a little tedious unless you can write some scripts to automate it. The Symantec Ghost "server" solution will also do what you want. –  Evan Anderson Jun 18 '09 at 15:34
    
Clonezilla isn't so hot. I prefer FOG. fogproject.org –  Wesley Dec 1 '09 at 23:35
add comment

Windows 7 will be great for this - as I understand it the native backup software will backup to a virtual disk that can be booted against new hardware.

share|improve this answer
    
Vista will backup to VHD as well, but I don't believe that it can currently boot from a VHD. –  Evan Anderson Jun 18 '09 at 14:25
    
Yes you're right, it won't boot. The new Win7 format will boot and can run against new hardware (although i've not tried it myself) –  Chris W Jun 18 '09 at 15:45
    
I'm booting my laptop off of VHD using builds 7000 and 7100 of Windows 7, and I really like it. I ran build 7000 for quite awhile on the bare-metal NTFS, and I don't see a noticible performance difference running out of the VHD. It's quite nice. –  Evan Anderson Jun 19 '09 at 3:25
add comment

wow, that's a good problem.

So if your requirements are near-instantaneous recovery of what are essentially dozens or hundreds of "one off" installations...then you need nightly backups onto a centralized server which creates installable images from the backups.

Bare metal restores are going to be your only solution for this, and to make it so that you don't kill yourself the first time this happens, all of the laptop models should be identical.

It sounds like there's an excellent chance that your environment is not built to handle this sort of thing. There's not much you can do about that now, but I urge you to switch from an ad-hoc environment to a managed infrastructure. It will make your life easier and you will sleep better.

share|improve this answer
add comment

I've had excellent success using terminal services. Users workstations are reduced to thin-clients and they simply remote desktop in to a virtual machine. That virtual machine consists of two files (.vhd and .vmc) which can easily be backed up when not running. Backing up while running is another story altogether.

share|improve this answer
    
+1 This is the way to go. Hardware failure on commodity hardware (aka user workstations) is an everyday occurrence in larger deployments. Plan on being able to replace the thin clients on short notice. Plan to have a solid, reliable, redundant VRDP server that can handle the load and fail over quickly if needed. –  Max A. Dec 1 '09 at 21:18
add comment

If its 100% important that the user needs to be back online in like 10 minutes in the event of a hardware failure and you do not have any sort of generic machine as you pointed out so bluntly in your edit. The only way I can see to do this is to give every single person 2 identical machines and then use ghost or something to image the machine over once a day (or maybe it can do real time, haven't looked cause I don't have a need for this). This would give you another machine and at most the lose of work since the last image, less if they save to a file server on a regular basis.

Alternate, take action to keep the most common failure parts around, RAID 1 the system so if a drive fails they keep working. Keep extra memory on hand to replace that when it fails, if these are white box systems the power supplies should be pretty generic between them so figure out the ones you have connector wise and buy a bunch of those.

Another alternate, every night run VMWare converter on every machine when they log out, this will create a VMImage of their computer. Have the nightly run of this replace the previous one. Have a number of really powerful "spares" sitting around with VMWare workstation pre-installed, when the users hardware files just mount the last VM to the VMWare workstation, switch to full screen and they are working again. Rebuild their primary machine and then pay for the program to switch it back from the VM to the real hardware

What is it you are doing that requires everyone to have their own one of machine? I really don't see how going with a quasi standard load and ensure files are kept on the server can't help you? If my laptop failed tonight, I'd would be 100% operational in about 3 hours after getting a new one.

share|improve this answer
    
Good ideas, although nothing close to a ready-made solution. –  Toto Jun 18 '09 at 17:49
add comment

I don't know if this is a 'good' solution but you could 'roll your own' by having these machines booting off of PXE and wake-on-lan - where normally they would boot into the full desktop system, but at an interval and time of your choosing, they would boot into a 'nix and do a full drive imaging via dd or other imaging program of your choosing. Upon completion they would shut down, you have your full HD image backup, and the next time your user comes in their PC is exactly as they left it.

The cons are pretty obvious, you would have to set up a whole infrastructure and write some custom code to do this, but the pros is that I believe it would solve your problem exactly.

I don't know of any commercial software that does this, so if anyone else does please chime in! :)

share|improve this answer
    
Also, a full clone every time would be problematic across sites. –  Toto Jun 18 '09 at 15:15
add comment

Not something that will fix your problem now, but we are looking at doing "blade" computers from HP. Old school dumb terminals are making a comeback. Basically its one massive server feeding several dumb terminals their desktop image, but processing takes place on the server. The user is sitting in front of a screen with a computer-esque box attached with a nic, usb ports, etc. When the dumb terminal breaks, simply swap it out and the server back end is all virtual so there is essentially no down time in the case of hardware failures. The last Belk store I was in had this set up at all the employee kiosks.

If I remember correctly, HP had a solution for heavy processing as well. I believe it even supported CAD processing!

share|improve this answer
    
Right, and I plan to look at their blades very closely. However, it won't help with recovery across sites. –  Toto Jun 18 '09 at 15:14
add comment

I don't believe in backuping desktops. I'd see standardized software and Roaming Profiles with Folder Redirection as a partial solution. If your desktop fails, you boot up the new one and get most (not all!) of your settings and data back immediatly. If you have Exchange, that should take care of Outlook. Roll out Vista, so you can easily deploy generic images to practically any hardware (with help on DriverPacks for example) instantly.

share|improve this answer
    
+1 - Yeah-- what he said. –  Evan Anderson Jun 18 '09 at 14:26
    
It's not about what you believe, it's about what the business requirements are. –  Toto Jun 18 '09 at 15:14
    
Why not tell us why the requirement is that everyone be on one of machines? –  SpaceManSpiff Jun 18 '09 at 16:42
    
@Toto: Funny thing-- I'd say the opposite. Your business requirements are wholly unrealistic (at least with today's operating systems and hardware), and rather amusing. You're tilting at windmills, basically. Your whole strategy flies in the face of having a PC replacement lifecycle with a remotely reasonable TCO, let alone trying to do DR. You'd be getting what you're asking for "for free" with a more sane PC state management regime. Your application and operating system state on these PCs cannot be so complex that its configuration couldn't be scripted. Good fun, this. –  Evan Anderson Jun 19 '09 at 3:33
add comment

Platespin Protect offers this functionality. It takes live copies of production machines, and is able to spin up a DR replica of the 'workload' on-demand. Replication runs in the background and can replicate across a WAN. It's source-agnostic, and supported targets include ESX, Hyper-V and Xen.

Time to bring a DR copy online is about 5 minutes. It boots the target VM, installs relevant device drivers into the OS, then brings the box up with networking settings that you specify. You still have to have some means of your end user accessing this DR replica, however.

The key issue you have here is your criteria that the replacement be brought online without any delay, while also being an identical copy of a highly customized machine. What this implies is having some kind of 1:1 permanent sync between the machines. This is clustering. You want to cluster desktops.

My gut feeling is that providing a VMWare VDI environment for your users will be the best solution for you. This makes the local workstation irrelevant as long as you can get it to connect to the VMWare infrastructure. They'll also be able to run their personal environment customized in whatever manner they like, and on the back-end you can provide proper redundant power, cooling, and solid VCB-based backups.

share|improve this answer
add comment

Symantec has a product from the BackupExec suite called System Recovery (previously LiveState System Recovery). This product is designed to create images from the system on the fly with no need to shutdown the computer.

You can set a low periodicity for the images (so you don't lose much data) and consolidate them afterwards to save space.

Images can be saved to a central location and can be restored in different hardware or even virtual machines.

In the case of desktops is not very expensive (which is not true for servers). When I looked into it there was only a Windows version (no Unix supported).

share|improve this answer
add comment

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.