Designing Distribution of Computations for Mobile Cloud Computing Systems Thesis proposal

A Mobile Cloud Computing (MCC) system is a cloud-based system that is accessed by the users through their own mobile devices. MCC systems are emerging as the product of two technology trends: 1) the migration of personal computing from desktop to mobile devices and 2) the growing integration of large-scale computing environments into cloud systems. Designers are developing a variety of new mobile cloud computing systems. Each of these systems is developed with different goals and under the influence of different design constraints, such as high network latency or limited energy supply. The current MCC systems rely heavily on Computation Offloading, which however incurs new problems such as scalability of the cloud, privacy concerns due to storing personal information on the cloud, and high energy consumption on the cloud data centers. I propose to address these problems by answering the following research question: “How computations should be distributed across different computing nodes in MCC systems?” For a quantitative analysis of mobile cloud computing systems, I have developed the first generation of an innovative design and simulation tool that offers large scalability and heterogeneity support. With this tool system designers and software programmers can efficiently develop, optimize, and validate large-scale, heterogeneous MCC systems. Leveraging this tool, I developed two new MCC systems where I applied two variations of a new computation distributing technique, called Reverse Offloading. By more actively leveraging the computational power on mobile devices, the MCC systems can reduce the total execution times, the burden of concentrated computations on the cloud, and the privacy concerns on storing personal information on the cloud. This approach also creates opportunities for new services by utilizing the information on the mobile instead of accessing the cloud. I will continue pursuing this line of research by investigating the following ideas: 1) further improving my tool by adding support for modeling mobile GPU systems and 2) developing new MCC applications with new variations of reverse offloading. The ultimate goal of my research is to enable the design of better mobile applications and cloud-computing platforms. In particular, it will become a vehicle to optimize not only their performance but also their energy dissipation, an aspect of critical importance for any computing system. This proposal is organized as follows. First, I describe the research problem, the distribution of computations in MCC systems. Second, I present NETSHIP, a design and simulation tool for MCC systems. Third, I describe a cluster of embedded devices and the reverse offloading technique I developed. Fourth, I present a local training MCC application for information extraction and another reverse offloading technique I invented. Fifth, I list related research projects. Finally, I propose a plan to complete my research with a tentative timeline.