| Smart mobile devices have increasingly become the computing platform of choice. As distinct features, they have sensors and many useful applications based on the sensors have been developed and are being widely used. Not surprisingly, many of these applications rely on the cloud because of the resource constraints inherent to mobile devices.;In this dissertation, we explore systems support for enabling the efficient processing and secure sharing of sensor data using the cloud. To achieve this goal, we adaptively use the cloud for mobile devices based on the availability of hardware components, networks, and the state of human workers. When appropriate, we exploit domain knowledge and use mathematical analysis to fulfill our goals and system requirements.;We first enable compute- and data-intensive applications on mobile devices. Some emerging applications, such as interactive mobile perception applications, are too slow to run on mobile devices due to both their high data-rate workload (real-time videos) and compute-intensive algorithms used (computer vision-based ones). Thus, we need to significantly improve the performance by offloading part of the application components to the cloud and parallelizing it if relevant. To achieve this goal, we conducted a measurement study on the factors that can affect the performance of the applications and developed a novel, lightweight, runtime that automatically and adaptively makes offloading and parallelism decisions for mobile interactive perception applications.;We next focus on serious privacy infringements, such as the leakage of photos and the use of algorithmic recognition technologies by providers, when using cloud-based photo sharing services. Since currently we are required to expose our data to the providers without any controls, it is impossible for users to prevent the providers from such mining activities. On the other hand, the providers perform useful processing on our data to make our experience with their system better, e.g., scaling images to support mobile devices with different form factors appropriately. In order to simultaneously achieve both privacy protection and processing benefits provided by the cloud, we develop an image encryption/decryption algorithm and an associated system that can transparently work with existing cloud-based service providers.;Third, we focus on crowd-sensing, a novel capability that harnesses the power of crowds to collect sensor data from a large number of mobile phone users. In such activities, it is often challenging for users to efficiently deal with labor-intensive sub-tasks, such as recruiting workers, giving incentives, etc. However, existing programming systems do not handle these concerns appropriately. Our domain-specific language and runtime enables crowd-sensing and provides significant automation for such tasks. Users only need to provide a high-level description of a crowd-sensing task. Then, the runtime automatically takes care of the rest.;Fourth, when sharing large volumes of sensor data using mobile devices, we always have energy concerns. Modern smart mobile devices have multiple wireless interfaces, such as 3G/EDGE, WiFi, etc., for data transfer, but there is considerable variability in the availability and achievable data transfer rate for these networks. On the other hand, many mobile applications are often delay-tolerant, so that it is possible to delay data transfers until a lower-energy network connection becomes available. we present a principled approach to trade-off energy and delay on mobile devices using Lyapunov optimization framework. The resulting algorithm can automatically adapt to channel conditions and requires only location information to decide whether and when to defer a transmission. |
