Pushing the Envelope of Mobile Computing: Improving Security, Energy, and Latency by Bridging the Gap between Analytical Modeling and System Design Open Access
Downloadable ContentDownload PDF
Recent research findings and reports have raised the concerns for mobile security vulnerability and energy inefficiency. Besides, the growing popularity of smartphones, along with the rapid emergence of new domains of applications also impose new requirements for energy efficiency and service latency. These long-standing issues and new challenges call for novel approaches to further push the envelope of mobile computing, which needs to be secure, energy efficient and fast. On the other hand, the approaches should alleviate developers or users, who are usually without specialized knowledge, from convoluted configuration work.This dissertation aims to improve three significant aspects of mobile computing - security, energy, and latency - by bridging the gap between analytical modeling and real system design. Each problem considered is first formulated as an optimization-based problem, then the proposed algorithms to the problems are implemented on real mobile devices and evaluated using real-world traces and applications.First, contemporary permission-based security mechanisms on mobile platforms are proven to be ineffective to protect users' private data. We present SARRE, a semantics-aware security rule recommendation and enforcement system, which employs statistical inference and collaborative filtering techniques to automatically assign security rules to information flows in the system, thus preventing information leakage. The SARRE system is prototyped on Android devices and evaluated with 1473 popular apps from Google Play spanning multiple categories, and 213 real-world malware samples. The results show that SARRE overall achieves 93.8% precision and 96.4% recall in identifying the event paths, and also the average relative error between rule recommendation and manual configuration is less than 5%, validating the effectiveness of automatic rule recommendation. A camouflage engine implemented to enforce the recommended rules demonstrates SARRE is effective to provide the fine-grained protection over private data with low performance overhead.Second, in cellular networks, tail states are designed for a tradeoff between energy efficiency and latency. However, the energy consumed during tail states becomes a huge energy drainer itself. On the other hand, current app marketplaces are increasingly dominated by apps relying on mobile advertisements for revenue. However, traffic patterns of ad modules are inefficient for cellular networks. We reveal the reason for the inefficiency, and propose our design of the first ad management framework that is fully compatible with existing ad ecosystem, aiming to improve the efficiency of contemporary ad libraries, in terms of energy and data usage. The ad-fetching decisions are optimized based on a novel energy accounting method we proposed based on Shapley Value, a popular cost/profit allocation method in game theory. We fully implement the framework on Android and show that the system and network overhead is almost negligible. Our system achieves up to 45% energy savings than existing policies, and is transparent to mobile apps and contemporary ad ecosystem. To further capitalize the promise of saving energy by ad prefetching in real-world mobile systems, considering several significant runtime factors, we make a novel use of Markov Decision Process to model the energy minimization problem for ad prefetching (EMAP), and propose an algorithmic solution to the EMAP problem. Further, we implement the algorithmic solution with our ad prefetching system. By replaying real-world user traces on Android devices, we show our proposed solution consistently outperforms existing On-Demand policy on Android by up to 59% in saving ad-related energy, while a simple Fill-Up-Buffer policy can be even 2 times worse than the default On-Demand policy.Finally, due to the proliferation of the Internet of Things and interactive applications involving big data analytics and media processing, the gap between ever-increasing user expectations and limited mobile device resources has made mobile edge computing a promising technique. We propose a new optimization horizon, Quality-of-Result (QoR) in mobile edge computing, and present our systematic optimization framework, MobiQoR, to trade QoR for improved response time and energy saving. In our framework, mobile workload can be divided, offloaded and distributively processed by neighboring edge nodes. The optimization in our design aims to minimize the energy consumption and the service latency by jointly optimizing the task offloading decisions and the selections of all edge nodes' QoR levels. The proposed MobiQoR is implemented on real mobile devices. Using representative face recognition and movie recommendation applications, evaluations with real-world datasets show that MobiQoR outperforms existing strategies by up to 77.0% for face recognition and 189.3% for movie recommendation. Further, we consider real-world edge environment consisted of multiple edge clients and edge servers, we address two problems unique in heterogeneous edge computing: the cost accounting and workload assignment. To determine the cost of each of units of workload being concurrently processed on an edge-device, we propose to model the problem as a multi-choice game, and use Shapley Value for cost accounting. With the total cost decoupled, we are able to use the distributive Hungarian algorithm to solve the workload assignment problem in an efficient manner. We adopt a hybrid manner for evaluations: we profile the costs (including energy and data transmission costs) using a real implemented workload offloading framework in an edge environment, then the cost profiles are used to drive the simulations. Results show that our policy of workload assignment guided by multi-choice Shapley value is able to consistently outperform the two other baselines: Random policy and the other policy of Hungarian optimization based on Even accounting policy. The advantage of our policy is further enlarged when the heterogeneity level of the network or computing resource in edge environment is increased. We also show interesting patterns of the joint effects of different resources' heterogeneity levels and the weighting factor between them, which provide useful inputs for edge resource optimization.