Data-Oriented Approaches towards Mobile, Network and Secure Systems

With the rapid evolvement of information science, data-oriented research has solicited a new philosophy for the future mobile network and security design, since it can not only encourage new designs achieving more efficient and reliable networks, but also pose new challenges towards security designs. In this dissertation, we propose four novel data-oriented designs or frameworks to prompt or calibrate the performance with respect to efficiency, reliability, and security. In the wireless domain, packet corruption and packet collision are two major threats that jeopardize the performance of a mobile network. To cope with the packet corruption, we propose the STAteful inter-Packet signaL procEssing (STAPLE) framework, which is an inter-packet oriented signal process design for wireless networking. STAPLE transforms the signal processing procedure into a lightweight stateful process that caches in a small-sized memory table physical and link layer header fields as packet state information. The similarity of such information among packets serves as prior knowledge to further enhance the reliability of signal processing and thus improve the wireless network performance. For the packet collision, we present a new design called comb decoding (CombDec) to efficiently resolve RTS collisions without changing the 802.11 standards. We observe that an RTS payload, when treated as a vector in a vector space, exhibits a comb-like distribution; i.e., a limited number of vectors are much more likely to be used than the others due to RTS payload construction and firmware design. This enables us to use sparse recovery such as compressive sensing to resolve RTS collisions. For the network security design, we revisit network tomography and inference from a data-driven perspective and discover that there are two vulnerabilities. The first one is of the measurement integrity. By taking advantage of this vulnerability, we develop an attack strategy, called measurement integrity attack, which not only destroys the measuring system, but can even mislead the system to scapegoat other innocent users. The second vulnerability is of the measurement confidentiality since network inference is able to leak network flow information without directly measuring it. To prevent disclosing the flow information, we find that random routing strategies are capable of hiding the flow information. Then, by leveraging the measurement data, we propose a new framework that can systematically study the behavior of different randomized routing protocols, and explore the fundamental reason why randomized routing strategies can prevent information leakage against network inference