Slashdot Mirror
New Method for Random Number Generation Developed
Science Daily is reporting that a German team has developed a new method of random number generation that they hope will improve security. "The German team has now developed a true random number generator that uses an extra layer of randomness by making a computer memory element, a flip-flop, twitch randomly between its two states 1 or 0. Immediately prior to the switch, the flip-flop is in a 'metastable state' where its behavior cannot be predicted. At the end of the metastable state, the contents of the memory are purely random. The researchers' experiments with an array of flip-flop units show that for small arrays the extra layer makes the random number almost twenty times more 'random' than conventional methods."
In Numerical Recipes for C they list several benchmarks for determining how good one random number generator is compared to another (based on various statistics measures) so it certainly is possible for one method to be more random than another. Read chapter 7 of that book for all the details you could possibly want on this subject (with references to even more information).
One way of generating a good random number in Linux is using /dev/random (which uses a hardware-based random signal as its source, I don't recall the details). However, it isn't fast enough for most applications, outputting only a few bytes per second of random information, although it can serve as a useful seed for other random number generators. Just run 'cat /dev/random > random_bytes.bin' to see its output.
I'm curious what rate random information can be generated using the method in the article. I'm presuming it's fast enough that an application could rely solely on this data without having to use it as a seed for a pseudo-random number generator. The question is how long does it take for the hardware to get to the state where its next value is unpredictable--in the case of /dev/random it's relatively long.
You're confusing Shannon entropy and true randomness. If you have a string of bits that are created by a process that is truly random but has a bias, it's easy to transform it into an unbiased (but shorter) string.
The problem with pseudo-random generators is that they're really not random at all: They're determinstic functions that map a seed onto a sequence of random bits. If you know the function and the seed, you can predict all of it, which leads to potential vulnerabilityies. The point of truly random numbers is that there's no possible information you could have that would enable you to predict it.