I’ve been mulling over the same issue – but without the nice graphics -when it comes to desktop and application virtualization, and I’m very keen to dig a bit deeper into the user experience impact of a collection of new technologies:

1. Virtualised applications add a small performance overhead
2. Streaming virtualised applications adds a significant overhead to launch time, especially in a VDI environment where caching is of limited value (although pre-caching in the image would be better)
3. Virtualization of the applications configuration and the users personalised settings adds a further overhead to launch times
4. WAN access to data adds a further overhead to application launch times
5. We’ve yet to quantify for many niche applications whether non-persistent VDI images (where only the roaming profile is persisted at logoff) are going to be slower, maybe because they cache for performance in the local profile and assume that the users local profile is going to be there tomorrow 99% of the time
6. Sharing server resources across many users, is likely to work out great on average, but I’m not 100% sure that it will be faster for peak CPU periods which often occur at application start-up
7. Most VDI deployments encourage users to logoff frequently and that’s likely to increase as the logon/logoff cycle is required in order to update the master image, not only does this affect a few of the points above, but it also makes detailed user state preservation very important – ie saving which applications, files, scroll locations, browser tabs, window positions etc the user has open and restoring them when the user logs back in. 
8. I dread to think how regular logoffs would impact my productivity, right now I logoff once ever couple of weeks, and it takes me at least 20 minutes to close everything down and open everything up again, if I had to do this every day – the least of my worries would be the time it takes for the OS to boot.

So one things for sure, in the new word of desktop, end user experience performance monitoring is going to be pretty important.

What does the maze look like?  Well I’m not able to talk in detail about that right now, but Brian Madden sums up the situation well:

Ten years ago it was easy. We only had two options: Citrix (the new way), and traditional installs (the old way). But in 2007 we have application streaming, virtualization, and isolation; VDI- and Terminal Server-based server-based computing; local installations, local streaming, remote desktops, remote isolation, OS streaming, OS isolation…. the list goes on.

The one question I’m asked again and again is, “With all of these different methods of deploying apps, how do we figure out what we should use in our environment?” (The second most frequent question I get is, “what’s real and what’s not today?”

We too are trying to answer these questions, not just on a per application basis, which is the focus on Brian’s article, but for a customers entire application portfolio often numbering several thousand applications.  We’re also trying to ensure that our focus is not just on a point in time, ie we are trying to lay out a roadmap for change, that reflects the business, web 2.0, enterprise 2.0 and consumerization trends as well as the improvements in legacy application delivery technology.

We’re doing the work in the wiki that I posted about yesterday.

I’ve done quite a few enterprise operating system upgrade projects, and one thing I’ve learned is that whenever I started one of these projects I always pushed for the latest operating system available:

1. NT 3.51 instead of Windows 95 (I think) , this was considered a very big risk
2. NT 4 instead of NT 3.51, not so much of a risk
3. Windows 2000 instead of NT 4, this was considered a big risk
4. Windows XP instead of Windows 2000, this was considered a very big risk at the time

and always got the same pushback:

1. Wait for service pack 1
2. Driver supports terrible
3. None of our applications will work, and even if they do what about vendor support
4. Who will support it, no ones trained
5. etc

However what people always seem to forget is that enterprise deployments take a long time to plan and execute and that the world moves fast.  In every case mid way through the deployment it seemed unimaginable that we could have been deploying anything but the latest version and all of the issues had faded away. 

I think it’s the same with Vista, my view – get on with the planning, it will take you longer than you think and in 9 months time when you start volume deployments you won’t understand what all the fuss was about.

That said – please don’t think of your programme as a Vista upgrade.  Instead:

1. Model your workforce in terms of their workstyles
2. review the appropriate solution for each workstyle, looking at the many highly differentiated options for desktop and application delivery including consumerized/Linux solutions for some user segments
3. rationalise down to a few desktop/OS and application delivery technologies (not one per workstyle) to keep control of cost and complexity and increase flexibility
4. try and leverage the change programme to achieve some significant business improvement, cultural change and productivity improvements
5. Architect your solution to loosen the coupling of the desktop operating system to the applications, to make future change easier.