分布式系统测试关键技术研究

发布时间：2018-03-27 18:46

本文选题：分布式系统　切入点：测试模型　出处：《西安电子科技大学》2013年硕士论文

【摘要】：近年来，随着计算机以及网络相关技术的飞速发展，其应用领域不断扩大，业务逻辑越来越复杂，因此越来越多的网络应用采用分布式系统的方式进行构建。鉴于分布式系统的并发性、缺乏全局时钟和故障独立性等特征，，很难将单模块测试方法应用到基于分布式系统的模块测试中。分布式系统测试过程中主要存在两个问题：第一是难以保证分布式系统测试结果的可靠性；第二是在进行大规模分布式系统的性能测试及其它系统级验证时，存在着搭建分布式系统困难、测试效率低等问题。本文在对上述问题进行深入研究和探讨的基础上，分析了当前主流的测试模型，提出了一种改进的通用测试模型。此外，在该通用模型的基础上，提出了一种新的集群可靠性测试模型和一种新的集群性能测试模型，并设计和开发了用于支持和实现上述模型的原型系统。为了支持该通用模型，本文在研究了网络通信、Libevent、多线程以及异常注入等技术的基础上，使用Libevent框架进行事件处理，并采用多线程技术，优化了获取大量真实数据源时的性能，提供了高性能服务；设计并实现了通信链路接管方案，通过注入通信异常的方式，模拟了链路中的异常情况，为分布式系统可靠性测试提供了解决方案；提出了通信链路数据录制-回放方法，模拟了大规模的后端集群，解决了难以获取大量真实数据源的问题，为分布式系统性能测试提供了解决方案。实验结果表明，原型系统在面向搜索引擎的分布式系统测试过程中，取得了很好的效果。通过使用该系统的录制回放功能，达到了模拟集群的目的，使得测试过程中不再需要搭建集群环境，进而节省了大量的机器资源；通过使用该系统的链路模拟功能，成功模拟了模块之间的连接、读写等通信异常，从而保证了集群容错机制的健全性。在后续的研究工作中，将进一步加深对分布式系统测试关键技术的研究，并在此基础上，实现集群服务化和测试服务化。
[Abstract]:In recent years, with the rapid development of computer and network related technology, its application field is expanding, and business logic is becoming more and more complex. Therefore, more and more network applications are constructed by distributed systems. In view of the concurrency of distributed systems, the lack of global clock and fault independence, It is difficult to apply the single module test method to the module test based on distributed system. There are two main problems in the process of distributed system testing: first, it is difficult to ensure the reliability of distributed system test results; The second is the difficulty of building distributed system and the low efficiency of testing in the performance test and other system level verification of large-scale distributed system. On the basis of deep research and discussion of the above problems, this paper analyzes the current mainstream test model, and proposes an improved universal test model. In this paper, a new cluster reliability test model and a new cluster performance test model are proposed, and a prototype system for supporting and implementing the above model is designed and developed. Based on the research of network communication technology such as Libevent, multithreading and exception injection, this paper uses Libevent framework to deal with events, and optimizes the performance of obtaining a large number of real data sources by using multithreading technology, which provides high performance service. This paper designs and implements the communication link takeover scheme, simulates the abnormal situation in the link by injecting the communication anomaly, and provides a solution for the reliability test of the distributed system, and puts forward the method of recording and replaying the communication link data. A large scale back-end cluster is simulated, which solves the problem that it is difficult to obtain a large number of real data sources, and provides a solution for the performance testing of distributed systems. The experimental results show that the prototype system has achieved good results in the process of testing the distributed system oriented to search engine. By using the recording and playback function of the system, the purpose of simulating the cluster is achieved. In the process of testing, it is no longer necessary to build a cluster environment, which saves a lot of machine resources. By using the link simulation function of the system, it successfully simulates the communication anomalies such as the connection between modules, reading and writing, etc. Thus, the fault-tolerant mechanism of cluster is guaranteed to be sound. In the following research work, we will further deepen the research on the key technologies of distributed system testing, and on this basis, we will realize cluster service and test service.
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP316.4;TP311.53

【参考文献】