Killer Operating System is quite a nebulous concept. The precise things we want in a killer operating system change frequently, which means that it is probably doomed stay in the realm of our imaginations. Nevertheless, it is a great inspiration.
However, being all things to all people is hardly a good idea for most things - operating systems included. (Worse Is Better). It may or may not be possible to combine all Operating Systems Design Principles. Some people go so far as to say that instead of Killer Operating System, we should Kill The Operating System.
A killer operating system would hardly be killer without appealing to the masses - it must have a Killer User Interface.
And appealing to the masses is not enough for us programmer-types - it must have the things we like: Transparent Persistence, Killer File System, and Killer Reliability.
There is an enormous wish list at New Os Features.
A killer Operating System would implement All Data Relates To Other Data and therefore interface with a Three Dimensional Visualization Model to a Unified Data Model. Applications, instead of being monolithic, closed apps would be modular, loosely-coupled mash-ups that work with User Data and the Internet As Cloud Storage. The user-machine becomes used less as a computer, and more as a terminal, because the greater data is out there.
The relationships should be reactive, too, i.e. such that a change or repair in data in one location propagates to all the other data related to it. I expect synchronous concurrency will be necessary, to formally address the diamond-shaped dependency graphs and conditions where many things change at once or to avoid missing critical intermediate values. And we'll probably need to model latency relationships logically if we are to scale beyond a spreadsheet or integrate effects.
We've already seen a Killer Operating System: Sky Net. ;)
Don't forget Hal and GLaDOS
See original on c2.com 