Study Of Communication Real-time Performance In Distributed Embedded System

Ethernet has been extensively applied to the distributed monitoring field because of its good properties such as opening characteristics and low cost. As the embedded systems are widely applied to real-time monitoring field, it becomes a main development trend of embedded system to connect various kinds of embedded equipments to the Internet to realize the remote real-time monitoring.With the development of the Internet, TCP/IP protocol has become a de facto standard. Although the traditional TCP/IP protocol has put great efforts to guarantee the reliability of data transmission and control the data flow, it does not consider much for real-time system. Usually there are fewer resources in embedded equipments, however higher real-time performance is required in distributed monitoring system. So the traditional TCP/IP protocol can not satisfy the real-time requirements. Therefore, how to improve the real-time performance of embedded system communication becomes a very important and urgent problem t to be solved.The real-time characteristics in embedded system communication are not only referred to the real-time problem in an embedded system itself, but also to the real-time problem of communication networks and applications when various embedded equipments are interconnected. So it includes the real-time guarantees in both protocol and application aspects. This thesis is based on the background of the following two projects: Science and Technology Key Project of Shandong Province "Research on marine environment monitoring online and calamity intelligent early warning system" (Grant No. 2004GG2205108) and "Research and development on embedded system with independent intellectual property right" (Grant No. 2005GG1101001). In this thesis, the real time performance of distributed monitoring system composed of various embedded equipments is discussed systematically from theoretical and practical aspects.The main contributions of this thesis are included as follows:(1) A framework for embedded distributed real-time network is designed, and an embedded gateway is introduced into the framework. Based on the analysis on the new running mode and the low time and space complexity of the embedded gateway, a real-time database is then introduced and embedded into the gateway. At the same time, an M/G/1/ ∞queuing model is proposed for the access to database. In addition, a method combined with multi-attribute of data packets is presented to set the priority of packets. Based on the above, a simple but effective priority scheduling algorithm is proposed. The theoretical analysis shows that this model can effectively guarantee the real-time requirement of packets with high priority and fairness for packets with low priority. Moreover, it can reduce the queue length, is simple to control and occupies small amounts of memory and buffer space. So it is very suitable for the communication module of embedded internetworking equipments.(2)The embedded gateway includes limited resources, but requires high real-time performance. So the traditional TCP/IP protocol stack is not suitable for it. This thesis presents a new TCP/IP protocol stack for embedded system.In order to guarantee a good real-time performance of embedded system even the communication traffic increases, by analyzing various existing congestion control algorithms, this thesis proposes an improved TCP congestion control algorithm for the embedded system. By adjusting buffer size dynamically and dispatching data packets based on a two-dimensional table, this algorithm improves the transmission performance, the TCP communication capacity under heavy load condition and the real-time performance for important packet.(3)In view of the properties of embedded environment such as deficiency resources and special-purpose function, combined with Markov decision process theory, a finite horizon model is proposed to resolve the routing problem. By modifying the backward recursion iterative algorithms of Markov finite horizon model, a routing algorithm-Markov Finite Horizon Decision Routing (MFHDR) is proposed. Its correction and low time and space complexity are proved theoretically. In addition, the Markov Finite Horizon Decision Routing with QoS characteristic (MQoSR) is also proposed. The computation and topological information of these algorithms are distributed to all equipments, and each equipment works parallelly during a single phase. Therefore the work intensity of single embedded equipment can be reduced and balanced. This can make full use of each computing resource in the embedded "peer-to-peer" network.(4)To solve the problem distributed monitoring environment with large volume data, combined with theories in data mining field, fuzzy clustering theory and algorithms based on fuzzy relations are studied. In classic FCM (Fuzzy C-Means), the initial cluster centers are selected randomly. This sometimes causes the global optimal solutions unable to be found. In order to solve the problem, a modified FCM algorithm is first proposed based on the preprocessing of similarity relations between samples, which makes a good trade-off between computing reduction and algorithm stability. Then, to the problem of incomplete data clustering, an improved fuzzy c-mean clustering algorithm IDFCM is proposed. This algorithm considers the proportion of incomplete data in dataset and their effects to the cluster analysis. It solves the problem which classical algorithms can't be directly applied to. Finally, simulation experiments are carried out on the network system for real-time ocean data monitoring and transmission. The results show that these algorithms can not only reduce network data flow, but also improve the accuracy and real-time performance of prediction.