Homeland Insecurity 10025
Now, we're getting somewhere -- I have personally -- and very, very reluctantly -- set up pretty stable Outlook-Exchange (and OutlookBynari) systems, but only...
I'm not sure this story should be used as an example of why to switch --from-- Linux. My comments are marked with **. Linux may be nibbling away at Microsoft's Windows footprint...
Our customers were the CIA, NSA, FBI and lots of other 3-letter government agencies. The software was required to pbutt certain certification standards (FIPS, CMVP, etc). So the short answer is Yes, the output was tested and verified.
It works because the scheduler runs on an interrupt but it isn't all that precise. Each thread is in a tight loop meddling with the bits. When the scheduler switches a thread in or out the thread will not have run the same number of iterations each time. The number of iterations will vary by plus-minus several thousands of iterations each time the context switch occurs. It's neither controllable or predictable. I'm not the one who designed it.. but I have seen the code.
It's also worth noting that this isn't a "fast" algorithm. It's actually very CPU intensive to generate the random value so it's not something you would call to populate an array of 1-million random numbers. But if you needed a random number to generate a cryptographic key for a new account then this is what would get called.
If we could count on some form of "user input" it would have been easier. The delays between mouse movements or keystrokes contain a good degree of randomness. But our stuff had to work without the user pressing or doing anything.
Linux Advocacy from Newsgroups