Research And Implementation Of Fore-Processor In Distributed SLA Measurement System

SLA-Service Level Agreement, is a kind of contract between ISP and its customers for QoS purpose , in which both expected behavior and QoS parameters of service are included. Nowadays, more and more critical business activities are becoming INTERNET depended. This makes the requirement from customers for SLA increased. On the other hand, SLA can also help ISP to understand customers' requirement and maintain a good relationship with them. It can give an ISP a better position during the competition if SLA is available.Parameters negotiated in an SLA and the ways of measuring for getting them are the key points of SLA management and monitoring. A National 863 program project "Distributed SLA Monitoring System Based on Sampling Measurement" aims to design a prototype system "PERME" which can measure end-to-end SLA parameters on high speed networks using sampling method and passive measurement. As an independent third part, the system can provide customers and ISPs the values of parameters in their SLA contracts which actually happened. In PERME, all measurement nodes are self-controlled and have the same structure, which include a sampling monitor and an analysis server. An federal way to negotiate parameters is used between nodes for control when needed. An SLA analysis server consists of three parts: a fore-processor, a post-processor, and a web-based data interface. The thesis concentrates on the design and implementation of the fore-processor, and gives out the results of key problems' study on the way to the total solution, which are packet classification, buffer management, real-time computation of one-point metrics.The first chapter gives a brief introduction of the study background including the concept of SLA, the technology of network measurement, and the structure of PERME. The second chapter analyzes several algorithms of packet classification, and designs the classifiers for SLA monitoring and packet type statistics according to the requirements of PERME. The third chapter compares the circular buffer with the double buffer, and chooses the circular buffer as the buffer management mechanism of the system based on the results of the experiment. Metrics are parameters describing the performance and reliability of the networks. And flow is the foundation of metric computation. In chapter 4, four types of flow in PERME are defined. Then the definitions and computation methods of four one-point metrics are given. These metrics are response time, retransmission, misorder, and throughput. The fore-processor shall receives the data, classifies the packets, maintains the flow, stores the original data, and computes the one-point metrics in real time. A detailed introduction of the design and implementation of the system is given in Chapter 5 from the aspects of system structure, data flow, function modules, and etc. The sixth chapter designs the testing environment and scheme of PERME, and verifies the functionalities and stability of the system via static and dynamic tests. The last chapter summarizes the work of the thesis and prospects the future of PERME.