Research And Application Of Hotspot-tracking Technology Based On Virtual Machine JIT Mechanism

In recent years, Mobile Internet is gradually penetrating into people’s lives and work in various fields. A variety of applications with rich functions are growing rapidly, which is profoundly changing the pattern of people’s life in the information age. However, for the current mobile devices, mobile applications will always be restricted by computing resources and storage resources, thereby affecting the user experience. For this reason, it is necessary to migrate the time-consuming parts of applications to the cloud to run. By using the powerful resources in the cloud, applications will be accelerated greatly, while reducing workload of the mobile terminal. It is obvious that computation partition is one of the most important problems in the migration system. However, for the conventional fixed division, it is complex, coarse-grained and not flexible.In order to solve the above problem, a creative solution is designed and implemented to dynamically determine running hotspots of applications so that the division can flexibly adapt to the current operating environment. In this way we can confirm the optimal migrated objects, thus maximizing the efficiency of program. Because android is an open-source operating system and used widely, this paper conducted research work under the android platform. The focus mainly includes the following two aspects:First, because methods of accessing specific local resources or sharing local state can not be migrated to the cloud, this type of method needs to be filtered out before performing the real hotspot analysis. The paper proposed and implemented the meta-information analysis based decompilation of bytecode to extract relevant metainformation, such as call relations, modifiers and prototype, etc. After obtaining the meta-information, a noncomplete breadth-first traversal algorithm designed in this paper would remove all specified methods. When this was done, a result set would be imported to the application for use of subsequent analysis.The second research emphasis is to analyze and extend the partial implementation of virtual machine execution subsystem to achieve the goal of dynamically tracking hot functions. Unlike traditional dynamic analysis, the plan proposed in this paper needs no instrumentation, and needs no specific rules to write programs, even needs no aid of any analysis software. By thoroughly analyzing the operating principium of just-in-time compiler and extending it, then we could obtain the instruction-level complexity of methods. At the same time, we also expanded interpreter to obtain methods’ execution time and the current system load. After that, we would compute comprehensively all analytical results to determine hotspots of programs.The solution that was put forward in this paper reached a seamless integration between hotspot analysis and virtual machine execution subsystem; it not only solved much shortcoming of the traditional dynamic analysis, but also had great significance for practical applications. Finally, through experimental analysis, we further validated the accuracy and feasibility of the scheme.