Doctor of Philosophy (Ph.D.)
Random Number Generators (RNG) find use throughout all applications of computing, from high level statistical modeling all the way down to essential security primitives. A significant amount of prior work has investigated this space, as a poorly performing generator can have significant impacts on algorithms that rely on it. However, recent explosive growth of the Internet of Things (IoT) has brought forth a class of devices for which common RNG algorithms may not provide an optimal solution. Furthermore, new hardware creates opportunities that have not yet been explored with these devices. in this Dissertation, we present research fostering deeper understanding of and enrichment of the state of randomness within the context of resource-constrained devices. First, we present an exploratory study into methods of generating random numbers on devices with sensors. We perform a data collection study across 37 android devices to determine how much random data is consumed, and which sensors are capable of producing sufficiently entropic data. We use the results of our analysis to create an experimental framework called SensoRNG, which serves as a prototype to test the efficacy of a sensor-based RNG. SensoRNG employs opportunistic collection of data from on-board sensors and applies a light-weight mixing algorithm to produce random numbers. We evaluate SensoRNG with the National Institute of Standards and Technology (NIST) statistical testing suite and demonstrate that a sensor-based RNG can provide high quality random numbers with only little additional overhead. Second, we explore the design, implementation, and efficacy of a Collaborative and Distributed Entropy Transfer protocol (CADET), which explores moving random number generation from an individual task to a collaborative one. Through the sharing of excess random data, devices that are unable to meet their own needs can be aided by contributions from other devices. We implement and test a proof-of-concept version of CADET on a testbed of 49 Raspberry Pi 3B single-board computers, which have been underclocked to emulate resource-constrained devices. Through this, we evaluate and demonstrate the efficacy and baseline performance of remote entropy protocols of this type, as well as highlight remaining research questions and challenges. Finally, we design and implement a system called RightNoise, which automatically profiles the RNG activity of a device by using techniques adapted from language modeling. First, by performing offline analysis, RightNoise is able to mine and reconstruct, in the context of a resource-constrained device, the structure of different activities from raw RNG access logs. After recovering these patterns, the device is able to profile its own behavior in real time. We give a thorough evaluation of the algorithms used in RightNoise and show that, with only five instances of each activity type per log, RightNoise is able to reconstruct the full set of activities with over 90\% accuracy. Furthermore, classification is very quick, with an average speed of 0.1 seconds per block. We finish this work by discussing real world application scenarios for RightNoise.
© The Author
Wallace, Kyle, "Understanding and Enriching Randomness Within Resource-Constrained Devices" (2018). Dissertations, Theses, and Masters Projects. Paper 1550153802.