Research On Component-oriented Communication Middleware Technologies

With the application of parallel computing on science and engineering technology,people begin to study a simple and high performance parallel computing software framework.Because of the modules coupled to each other,the traditional parallel computing programs are difficult to maintain and reuse.Component-based software framework having the properties of high maintainability and reusability becomes the hot point of software research in recent years.This paper designs a portable,extensible and hierarchical component-oriented software framework for streaming data computing applications,and focus on studying the communication middleware technologies supporting the software framework.Firstly,this paper studies the mainstream parallel programming model and analyzes the MPI,OpenMP and Pthreads from the points of principle and programming model.On this basis,we design the hierarchical software framework.The framework can be divided into component layer,communication isolation layer and communication layer.The isolation layer can be further divided into publish-subscribe interface layer,publish-subscribe implementation layer and streaming data processing layer.Next we study the publish-subscribe model between components and design the pub-sub interface layer and pub-sub implementation layer.This paper implements two schemes of publish-subscribe model based MPI,the loop scheme and the group communication scheme.We study the principle of pipeline parallel computing deeply to implement the streaming data processing layer.At last,we select a typical signal processing algorithm to test our framework.The test result shows that the design of framework achieves the expected goal.No matter running on a dual-core computer or on a multi-core server platform,software system can take full advantage of the hardware's computing ability.The average utilization of CPUs all exceed 90%.The system achieves the requirements of high throughput and low latency.And with the improvement of hardware's computing ability,system's indicators have greatly improved.The portability,reconfigurablility and correctness of data have all been verified.