Looking beyond a singularity

How does one build a truly dependable software system? Seven years ago, Microsoft Research started the Singularity project to answer this question. The premise was to start with the best known software development tools and to build a new kind of operating system from the ground up. The operating system was to be both an output artifact and a laboratory for the research. Portions of the code and ideas have been incorporated into three separate Microsoft operating systems so far. I will give a brief overview of Singularity planned and built, then describe what we learned, both positive and negative. I will speculate on OS futures including current research to build an operating system in which every last assembly instruction has been verified for type safety, a system for truly mobile computation, and new tools for automatically restructuring large software systems.