Manageable Granularity in Mobile Application Code Offloading for Energy Savings

Abstract

Applications running on mobile devices bring users new and convenient experience. However, energy limitations on mobile platforms will reduce the performance and sustainability. Code offloading mechanisms partition application codes and offload partial workloads to remote servers for energy savings. However, code offloading effect tends to be challenged by network transmissions as desired responsiveness maybe violated. Thus, the decisions on partitioning granularity become pivotal. In this paper, we propose a two-phase partitioning mechanism called MACO which maintains both the application responsiveness and a manageable granularity. We find that transmitting UI information is inefficient in network communication. So application codes are automatically divided into computational and UI-related segments firstly, where the latter ones are preferred to run locally. Secondly, the programmer can fine tune the intermediate partitioning result with a visual tool. We implemented MACO in web browsing applications written in HTML5/JavaScript codes. Evaluations on both real-world and simulated workloads show it reduces the response time (and corresponding energy) by up to 91% while the granularity is easy to manage.