Dynamic Offloading of Web Application Execution Using Snapshot

摘要

Mobile web platforms are facing new demands for emerging applications, such as machine learning or augmented reality, which require significant computing powers beyond that of current mobile hardware. Computation offloading can accelerate these apps by offloading the computation-intensive parts of an app from a client to a powerful server. Unfortunately, previous studies of offloading in the field of web apps have a limitation for the offloading target code or require complex user annotations, hindering the widespread use of offloading in web apps. This article proposes a novel offloading system for web apps, which can simplify the offloading process by sending and receiving the execution state of a running web app in the form of another web app called the snapshot. Since running the snapshot restores the whole app state and continues the execution from the point where it was saved, we can offload regular web app computations that affect the DOM state as well as the JavaScript state, and we do not have to pre-install the app binary at the server. Moreover, the snapshot does not require any annotations to be captured, making computation offloading more transparent to app developers. We qualitatively compared the proposed system with previous approaches in terms of programming difficulty and the scope of offloadable codes. In addition, we implemented the proposed system based on a WebKit browser and evaluated the offloading performance with five computation-intensive web apps. Our system achieved significant speedup (from 1.7 to approximately 9.0) in all of the apps, compared to local execution, which proves the feasibility of the proposed approach.