Ebook Transparent Method Offloading for Slim Execution
Mobile devices increasingly pervade our life and become one of the necessities in our everyday life. Although recent mobile devices are more advanced than early versions, users and developers feel much more enhancements are still needed. Most of the users already have a good experience with their powerful, graphic rich desktop PCs. In comparison, all they can find in mobile devices are still primitive applications they once used in their PCs ten years ago. Worsening the situation, mobiles devices are only up a few hours due to the limitation in battery capacity.
PCs are now commodity products. You can find many PCs located in public places such as coffee shops, airports, hotels, and so on. In the near future, those PCs deployed in public will be a great leverage to the future of pervasive computing. In this paper, we propose transparent method offloading as one way to take advantage of PCs placed in public and overcome resource limitations in mobile devices. The principle of our approach is to execute the computationally heavy parts of applications on nearby powerful PC severs and the other parts on mobile devices. Existing work have similar approaches to ours. Applying application partitioning at class level, they can distribute heavy computations on servers, but they have to handle them at class granularity. In contrast, we can explicitly select heavy methods to offload instead of the whole class and all of its methods.
To verify our idea, we developed the Distributed Execution Transformer (DiET), which automatically modifies Java bytecodes to offload selected methods. Since Java becomes increasingly popular among mobile application developers, we believe Java applications are good candidates for the purpose of our study. In addition, we can easily analyze and transform Java class files, since Java bytecodes retains most of the high level structures of source programs. The DiET offloads computation-heavy methods by replacing their bodies with remote procedure calls. Since the transformation is automatically done, application developers do not need to worry about details of offloading. Moreover, the transformed applications are also legal Java bytecodes. Thus, they can run on normal JVMs without any special modifications, which saves extra efforts for users to install customized JVMs on their mobile devices.
The remainder of this paper is organized as follows. We begin by introducing the overview of the slim execution and the DiET. We next describe the details of our transparent method offloading. Then, we present our experimental results. Finally, we discuss related work and conclude this paper.
Download
PDF Ebook Transparent Method Offloading for Slim Execution
Posted in :
