Application Research Of Real-Time Linux In SMP System

According to Moore’s Law, when keeping the price, the number of transistors that can be placed inexpensively on an integrated circuit and chip performance doubles approximately every18months. However, this trend has growth slowing on embedded market now. The primary cause is the embedded processor besides considers the performance, but also needs to consider the factor of low power loss. The appearance of embedded multi-processor, especially on the multi-core or the multi-thread processor to satisfy the rapid growth of performance demand on embedded market. The symmetrical multi-processor (SMP) is the most widespread architecture in multi-core processor domain. But in the embedded real-time domain, there is quite deficient to the SMP research.Linux is the most widespread general operating system in embedded domain. Undergoes many year development, there has quite perfect support to SMP. In some soft real-time demand embedded market, due to the rich functions, like memory management, network support and so on, Linux also be popular.This thesis made a deep study in the3.0version of the Linux real-time scheduling mechanism. Above all these researches, it summarized some ideas of Linux real-time scheduling process, scheduling trigger, as well as balancing strategy of real-time scheduling in multi-processing architecture.Varis practices and theory are combined, taken the system of a video surveillance as example. This thesis made some changes in multi-threaded and real-time transformation ability of this system, and performance are significantly improved by testing results. Finally, aiming at the lack of parallelism of Linux, a fine-grain parellel optimizing method is adopted. The Intel Vtune performance analyzer is involved in the parellel analysis, in order to find hotspots from the encoder of surveillance system, then parallelism optimization. The experiment result shows that the performance of encoder is greatly improved by taking advantage of the parallelism of symmetric multiprocessing.