The Research And Design Of Performance Profiler For Embedded System

With the rapid development of computer hardware and software technology, the scale and the capability of hardware of embedded system has been greatly improved. Accordingly, the complexity and scale of system software and application software of embedded system also grow increasingly. Simultaneously, the particularity of embedded system asks for embedded software being more smart and reliable. There is no doubt that the R&D of software plays a more and more important role in R&D of the whole embedded system. The quality of embedded software plays a decisive role to the quality of the whole product. So the embedded system developers need a performance profiling tool designed for embedded domain to improve the quality and reliability of the embedded software.In this paper, we study the basic principle of the embedded software performance profiling, research and the analysis of the major embedded performance profilers and their technology route even though their implementation. We focus on a performance profiler named gprof which is developed by the GNU organization. On the basis of understanding the key technology of gprof, such as its working route, its collection for performance data, its performance data file and so on, we understand the advantage and the disadvantage of gprof. Then we force on the two aspects of the disadvantage which the gprof have, the collection and the handling of performance data. On the basis of the research before, we cut in the point of gprof. We digesting the advantage and focusing on the disadvantage of the gprof tools,make sure about the technology route and the requirement, design and implement a new embedded performance profiler called EM-Profiler which is in plugin mode.On the aspect of the performance data collection, we design embedded dynamic detect frame EM-DDF and embedded dynamic probe EM-DP to collect the performance data in low overhead, and we implement several kinds of probe to collect special data in the EM-Profiler. On the aspect of performance data handling in the EM-Profiler, first, we design a new embedded communication protocol to implement the data transferring in high performance between the host and the target, and we design a new communication module to implement the profiling between several hosts and several different structural targets. Second, we use the native Bayesian classifier to classify the performance data. Finally, we design a method called analytic hierarchy gray evaluation method to evaluate the performance of the whole embedded system. Then, we integrate the EM-Profiler in the Eclipse IDE as a standard plugin to improve the convenience, the Portability and the Extensibility. Furthermore, we analyze the performance of the EM-Profiler and use it in an embedded project. The result shows the new profiler is still a low overhead. It has little effect on system performance. The result proves the EM-Profiler has high practicality. At last, we summarize our main works and introduce our research works in the future.