The 5-year period between Windows XP and Windows Vista was a career highlight – partly due to what we accomplished, but also due to the great team that I got to work with. The LDDM (Longhorn Display Driver Model) team consisted of around 10 people and was led by Anuj Gosalia, and was further broken down by two sub-teams – Steve Pronovost led the scheduler and memory management effort and Sameer Nene led integrating the new infrastructure (which is where I worked). The extended team was very tight and worked well together.

As this effort required a significant amount of driver changes, Nvidia, AMD, and Intel had onsite representatives co-located near our offices. We also obtained the source code for AMD’s R200 driver, which allowed us to implement driver features as part of our operating system development.

The overall development story of Windows Vista was very chaotic. Windows XP was internally code-named Whistler (after the ski resort) and the next version was to named Blackcomb (after the same resort), but the Blackcomb plans grew too large, so they decided to make a much smaller release (somewhere between Windows XP and the Blackcomb plans). At the Whistler Blackcomb resort, the Longhorn saloon is situated between the Whistler and Blackcomb gondolas, so “Longhorn” became the internal codename of this scaled backed window release.

But of course, the scope of this scaled-back release slowly expanded and what was to be relatively quick release turned into a 5-year project – the longest and most chaotic Windows release ever (plagued by security issues/ training and a quality reset that effectively threw away almost 10 month’s worth of work). Microsoft was also very political with a great deal of empire building, and as LDDM was a very visible project, some leaders wanted it to fail. This only served to strengthen our team cohesion.

Our team designed an interface layer between the existing runtimes and the new kernel interfaces, and also designed large portions of dxgkrnl.sys. One of my contributions was to provide the first synchronization model to dxgkrnl.sys, where most of the time everything was very fast and streamlined, but other times we’d have to wait for all outstanding actions to finish (such as mode changes and device removal).

The success of the DirectX APIs made our job more difficult, as literally thousands of applications used DirectX (often relying on undocumented behaviors) and we had to ensure that every one of them ran reasonably well on the new architecture. We used the entire 5 years to complete the task, and we were very busy for the entire time. That entire period feels like a blur.

As I work in today’s post-covid (largely remote) world, I reflect on why it is so hard to sometimes accomplish much simpler projects, and I think that the key to our success was based on:

• We had great people on the team, including strong architects and developers and we tried to keep ourselves aligned.
• We had lots of high-bandwidth communication. We all worked in-person and our offices were collocated. We even had a small conference room that was entirely ours that could be used for impromptu discussions.
• We had great coordination with IHVs, which were our largest dependency. They were onsite and we all had direct access to each other.

Windows Vista was probably the most poorly received Windows release to date (mostly due to the chaotic planning and execution) – it was considered buggy and slow. Some of this was due to our work on LDDM, but any such issues were fixed in the next Windows 7 release.

I well remember when I finally installed Windows Vista on our home computer and was showing it off to my wife. I pointed out how the windows had a cool glass edge that blended with the content behind it. She said that it was nice, but she hoped that nobody had to work weekends to make that happen. It was demoralizing, but my teammates thought it was funny.