Exhausted

You're not talking different things.

At the moment, we don't know how to manipulate atomic-scale electrical noise. Presumably, something like an array of phonon lasers (sasers) might be able to manipulate or skew the statistics, and by that I mean, besides obvious high level changes, like increasing the temperature by bombarding the target with wideband incoherent acoustic energy, a.k.a. heat-as-we-know-it!

Which is to say, electrical noise is, on its most basic level, simply taking a huge enough convolution of sources (random directions, frequencies and amplitudes) that, even if perfect knowledge could be had of the phonons in the solid, the electrical signal resulting from those particles couldn't be predicted exactly within the age of the universe.

The same could be done with the Internet, but it's not quite large enough for that to be viable. It's still possible for a very powerful adversary (like the NSA) to record far more data than your net-RNG system collects, making it vulnerable.

When Moore's Law finally winds down towards its information-theoretic limit, quantum entanglement will become that much more elementary to "basic superultrahigh-bandwidth" communications, and it will finally become impractical to make copies of intercepted communications. At the moment, we're something like 10^6 times more power-inefficient than theory, so storing traffic wholesale is down in the noise floor (all it costs is our hot and bulky RAM and hard drives, and the power to run them.. so, er.. yeah).

Tim

Don't laugh - a distant colleague wrote a hash function that inadvertently did that, and it got shipped in a product (long time ago). It wasn't found until we were looking for an N**2 performance problem in a hash table which it keyed - although the base vector was being expanded, all elements were in the overflow chain of the first entry.

I know I've said this before, a 20V zener from 30V rail, with ~a few hundred k to meg of resistance, makes big noise spikes consistently (in the dark). Band width drops off above ~1 MHz

George H.

t,

superultrahigh-bandwidth" communications, and >it will finally become

, we're something like 10^6 times more power->inefficient than theory, so storing traffic wholesale is down in the noise >floor (all it costs is our hot and bulky RAM and hard drives, and the >power to run them.. so, er.. yeah).

And as the sun sinks slowly into the west, we bid a fond farewell....

Cheers

Phil Hobbs

I'll give you a random number... 7. Guaranteed to be random.

At higher currents, zener noise quits being spikey and starts to be Gaussian. Some zeners are flat up into 10s of Ghz.

The issue with real avalanche diodes is that they need a high voltage supply. In something low voltage (e.g. single Li cell or USB powered) this requires a boost converter or charge pump. Then you need to be careful to keep the switching noise out of the low amplitude "good" noise, assuming you haven't already ruled it out on the basis of board space or BOM cost.

Regards, Allan

Everything hashed to the same number? A hashed symbol table would still work... it would just get slow.

Warning, you are about to re-discover the famous Debian SSL bug:

Their seed was non-deterministic, but too predictable.

Best regards, Piotr

If the application is required to produce composite hashes (i.e. merge two or more hashes and produce a new one), the bugs are even more subtle. I've seen using a simple xor on the input hashes, which is theoretically correct (if a and b are random, then a xor b is too), but fails badly if the element come in non-unique pairs, which is frequent. Then the result of a xor a is not as random as one would expect... Also found using a profiler. :-)

Best regards, Piotr

Seems to me that truly random numbers are inherently uninitialized data. Isn't there a 'volatile' directive for stuff like this?

(Beginner here)

Say, did you mean that transistors get quieter as they age?

(intrigued)

Michael

When you reverse bias the emitter to base junction enough it breaks down and produces lots of random electrical noise, but over time the amount of noise decreases and after a long time (many months or years) it may no longer produce enough noise to be useful. It also ruins the transistor for anything else.

Digitizing screen shots of an aquarium would provide the same thing - just don't try to save on hardware by digitizing a screen saver.

That method was suggested by Shannon. He also predicted the bug you found. If the data is correlated than Shannon's method makes it worse.

It was random, until the moment you wrote it down. Now it is always a 7 as anyone who reads your post can affirm.

Oh! Okay. So much for the idea that over time, the transistor becomes better :p

Thanks for the clarification

Michael

To be fair, Eric never intended the default RNG to be used in anger. I know that, because it was he who told me that we should consider a better source of randomness, especially for for banking. And then the security auditors crawled over my implementation in some detail.

(Eric = Eric A Young, author of SSLeay, lately known as OpenSSL).

No, for me it is still random. ;-)

Best regards, Piotr