The ability to generate random numbers -- to flip coins -- is crucial for many computing tasks, from Monte Carlo simulation to secure communications. The theory of building such subsystems to generate random numbers is well understood, but the gap between theory and practice is surprisingly wide. As built today, these subsystems are opaque and fragile. Flaws in these subsystems can compromise the security of millions of Internet hosts.
This project will develop black-box techniques for finding entropy failures at Internet scale related to randomness. These techniques, which build on programming languages, operating systems, networking, security, and cryptography, will then be applied to perform a systematic, ongoing census of the state of random number generation vulnerabilities. Through this census, the project can analyze the "longtail" of deployed networked systems rather than a handful chosen a priori.
The project will develop and disseminate defensive analysis tools that will help programmers debug and correct entropy problems in their code before they ship it. It will produce cryptographic countermeasures and entropy-gathering subsystems that provide better security guarantees than existing systems, reducing the impact of those entropy failures that do slip through.
The result will be a better understanding of entropy and more trustworthy systems, today and in the future.
