Extracting randomness from peripheral devices.

Abstract

Random number generation is at the core of cryptography, as the strength of an encryption scheme lies in the unpredictability of cryptographic keys. Unfortunately, available randomness sources are not always robust. Current options for the GNU-Linux operating system include the base random device (-dev-random), the unblocked random device (-dev-urandom), and the Audio Entropy Daemon (which uses a microphone for bit generation). In this work, we propose an improvement on the aforementioned generators, using an extractor to obtain random bits from peripheral devices. We have considered three di erent extractors, namely Trevisan's extractor, hashing on nite elds and HMAC. We have also considered three di erent peripheral devices: microphones, hard drives and networking devices. Standardized randomness tests were used to validate the di erent techniques, namely the NIST Statistical Test Suite and the Diehard Battery of tests. HMAC had the best performance out of the extractors considered, with all three input devices. It also outperformed the Audio Entropy Daemon, but had a similar performance as -dev-urandom. Compared to the latter, our design had the added advantage of the resiliency to security attacks, which is a feature the current unblocked devices do not share.