LCE: Don't play dice with random numbers [LWN.net]

LCE: Don't play dice with random numbers

Posted Dec 3, 2012 17:23 UTC (Mon) by Kaejox (guest, #85586)
Parent article: LCE: Don't play dice with random numbers

Some programs seem to drain entropy fast. Also starting a new process needs random numbers from kernel pool. Your can watch how much is available at /proc/sys/kernel/random/entropy_avail
It's possible to plug in hardware based RNG Simtec Entropy Key (USB stick) that keeps entropy pool full, so existing systems can benefit from steady stream of good random numbers.
http://www.entropykey.co.uk/

to post comments

LCE: Don't play dice with random numbers

Posted Dec 3, 2012 18:32 UTC (Mon) by madhatter (subscriber, #4665) [Link] (1 responses)

I have no idea if you're involved with Simtec, but I'll add my 2p-worth with the disclaimer that I'm not, I'm just a happy customer who once bought one of the Entropy Keys at full price. I like it. They're honest about what's inside the device, which is itself an elegant thing, and it keeps my poor, previously-entropy-starved colocated server full of nice chewy randomness.

LCE: Don't play dice with random numbers

Posted Dec 4, 2012 15:46 UTC (Tue) by nix (subscriber, #2304) [Link]

Likewise here. It's useful for its stated purpose, and its design has that nice polished, 'we thought of everything' feel to it. (It's also a disproof of Bruce's comment that all you need is a single diode junction and appropriate detector: you want two and a correlation detection algorithm of some kind, and probably a thermometer as well, to protect against both particular known attacks (e.g. heating the thing up) and unknown attacks against the physical device (which would be likely to affect both diodes in the Entropy Key, thus causing some degree of unexpected correlation between the two). Even then, unknown attacks that bias both diodes yet cause them to remain apparently uncorrelated but actually correlated will still slip through.