Execution Offloading Techniques to Optimize Mobile Cloud Computing

Abstract

Smartphones and tablets are rapidly becoming the computing device of preference in the global internet connected device market. Following the trend of the time, the users spend more time on these smart mobile devices (SMDs) using highly sophisticated applications, such as vision, graphics and augmented reality. It is still challenging to deliver such complex applications on SMDs, however, due to the key resource constraint like limited battery and low network bandwidth. In order to tackle this problem, recent studies suggested mobile cloud computing techniques that attempt to connect resource-constrained SMDs to nearby resource-rich powerful clouds. These techniques often imply execution offloading (or computation offloading), which is a promising technique to effectively deliver mobile cloud computing into the real-world mobile computing environments.

The main purpose of execution offloading is to throw the computational burden of SMD to the powerful servers by migrating a process or executing a method remotely. To achieve this goal, the current application state is captured and transferred to the servers over the network at runtime in execution offloading. Expectedly, the state transfer cost for the application state is a deciding factor for the success of execution offloading

because the size of the application state may reach up to multi-megabytes at a time, reducing the transferred state size is very important to maximize the benefit of execution offloading. In this dissertation, I propose novel techniques based on compiler code analysis that effectively reduce the state transfer cost by transferring only the essential application state actually referenced in the servers.

Another observation for execution offloading is that the early offloading studies depend on many idle assumptions. For example, they assume that the performance of a target server is always idle and constant. In the real-world commercial cloud environments, however, the cloud provider tries to maximize the server throughput by running as many applications as possible on a single server (i.e., oversubscription) and it makes such assumptions unrealistic. To design more realistic offloading scheme for the real-world cloud environments, therefore, it is necessary to consider the cost-effective behavior of the cloud platform. In this dissertation, I introduce a new cost-effective execution offloading scheme, called CMcloud, which not only maximizes the server throughput but also satisfies the post-offload performance of all target applications.

One challenge in execution offloading is to design the application-specific offloading techniques. Many mobile applications have their own, unique characteristics and some of them may make the strategy of the existing studies fail. It is important to adopt target-specific optimizations into offloading framework, therefore, to improve further the performance of target applications via execution offloading. To show the opportunity to achieve this goal, I suggest a streaming-based execution offloading framework that successfully guarantees quality of service (QoS) of 3D video games. I further propose live offloading, which allows transferring the current application state before the remote execution of the offloaded application actually begins, to make the suggested framework even more effective for better user experience.