Research On Virtualization For The Mobile Code Offloading

As large numbers of mobile applications become more complex, phone's demand for computing resources is increasing. Because cloud can provide unlimited computing power, storage and electricity, the way of offloading mobile applications to the cloud for remote execution can improve the phone's computing power and increase the durability of mobile phone batteries. However, the existing cloud systems won't start the virtual machine until a request comes, which is called fragmented management, and just allocate one virtual machine for a request's serialization execution. This results that each request needs a long response time, which severely affects the user experiences. Therefore, the research on virtualization for the mobile code migration has important theoretical and practical value.The key of research on virtualization for the mobile code offloading is how to improve the existing system's fragmented management and the serialization strategy to reduce the request's response time. First, by encapsulating terminal commands of the virtual tool, we achieve a clone pre-start mechanism, successfully replacing the fragmented approach, which not only provides the underlying support for pre-start algorithm, but also reduces requests' average waiting time. Second, based on the division of clone's valid states and the definition of control commands, we combine timing detection of the main program with real-time tracking of the control flow successfully to achieve the output of clone's state information, which is good for the design of clone's further management. Besides, this study puts forward the concept of primary clone and secondary clone and use different dispatch strategies to achieve allocation on demand of resources and high parallelization execution of the requests, which effectively reduces the cloud's response time.The experimental data shows that: the proposed strategies are especially effective for some typical applications. The results of the N-Queens solver, virus scanner and face detector indicate the response time of paralleling execution reduces by 35.60%, 38.71% and 30.08% compared with those of serialization.