Dynamic Remote Resource Utilization Mechanisms on Mobile Computing

Abstract

In the era of the Internet of Things (IoT), mobile computing is no longer limited to local resources. The rapid progress of mobile wireless networks such as WiFi and LTE enables mobile users to easily access various remote resources (e.g., CPU, DRAM, peripheral devices) in different devices (e.g., desktop, cloud server, smart TV). As a result, such an environment has enabled a lot of interesting services utilizing remote resources dynamically, which is infeasible in a single device alone due to its resource constraints. For example, one mobile user can offload compute-intensive tasks (e.g., face recognition) to cloud servers to improve performance and battery life. Also, another user can simultaneously utilize remote I/O resources (e.g., screen, camera) located in different mobile devices to construct a user-customized service. However, existing system software on mobile devices has a limited capability to support such services. Although some solutions can provide the functionality to access remote resources, they have suffered from severe performance degradation and inefficient resource utilization. Moreover, as mobile devices access remote resources using a slow and unreliable wireless network, naive system software design often exposes the problems more seriously. In this thesis, we propose the mobile computing framework to efficiently and dynamically utilize remote resources from possibly different devices and servers. The key idea is to identify performance-critical resources and possible performance bottlenecks in advance, and to timely redistribute tasks (or resources) across devices and servers. To achieve this, we leverage three different techniques to utilize cloud resources and mobile resources. First, we maximize the throughput of computation offloading in cloud servers. We estimate post-offload performance and perform load balancing for offloaded tasks while minimizing resource contentions. Second, we increase the duration of application caching in a mobile device. We extend a device’s memory capacity and timely prepare essential memory pages using the cloud-assisted memory swap mechanism. Lastly, we improve the performance of I/O sharing between devices for multi-device services. We seamlessly integrate multiple resources in different devices and adaptively redistribute tasks between devices. The evaluation results show that our framework outperforms conventional software solutions regarding resource efficiency and performance.