Layered Queuing Network Modelling Of Database Systems In Cloud

Thanks to the rapid advance of the information technology and its industry from the late twentieth century to the beginning of this century, cloud computing is now the fifth utility (after water, electricity, gas, and telephony). Cloud computing is that all kinds of computing resources including hardware resources and software resources that can be located in the local or remote can be connected by Internet in order to make up a virtual computing resources pool. It provides services through the unified management and operation for the needs of cloud users.Cloud computing is a calculation model with high efficiency and low cost. It not only creates considerable profits for cloud providers but also provides significant conveniences for cloud users. Cloud providers can take advantage of bulk purchase and management of a lot of computer clusters and optimally design them so that the resource utilization is enhanced, while reducing costs. For cloud users, cloud computing is simple and flexible, which can free them from the processes of procurement, installation, management and maintenance of computer software and hardware, speed up the application deployment, and reduce the cost of the configuration and the management of computing resources.In cloud computing, database systems are deployed in virtual machines as a virtual application. Cloud users make use of the physical resources in terms of virtual machines. In the case of cloud services, the performance of database systems is the focal spot that both the cloud providers and the cloud users pay close attention to. On the one hand, the cloud providers aim to meet the performance requirements of user application by means of employing fewer resources, which maximize the resource utilization and minimize costs. On the other hand, users aspire to the optimized performance, such as short response time and high throughput, of their applications with the fixed resources. As a result, in the course of controlling the configurations of the virtual machines in which database systems run, how to make appropriate decision of resources allocation is a key problem.As to the above issue, the forecast and the analysis of the performance of database systems is inevitable. This thesis studies how to establish the performance model of database systems by means of layered queuing network modelling in the cloud computing environment and solve the performance model by means of the method of layers, and explores the experimental environment which is built using Apache Hadoop to validate the performance model, which is used to allocate computing resources to the virtual machine in a given application of database systems.The main content of this thesis follows. Beginning with studying cloud computing, Hadoop, performance evaluation and performance monitoring method, this thesis researches how to establish the performance model of database systems by means of layered queuing network model in the cloud computing environment, probes into the workload of the database systems and its description method to determine the hierarchical relationships of workload requests of database systems, and discusses the input and output parameters of the performance model and the methods that obtain the input parameters of the performance model. In order to solve the performance model, this thesis explores the solving algorithms of the closed queuing network model and the open queuing network model and layered queuing network model, and studies how to solve the performance model that is proposed through the method of layers and predicts the average response time of database systems. This thesis also studies how Linux, Hadoop, Hive, HBase, Zookeeper and Ganglia can be used to set up the test environment.Under Hadoop, Hive is employed to carry out the query operation on database systems. The average response time of database systems are measured with both single user and multiple users separately. This thesis makes concrete analysis and comparison of the difference between the predicted response time and the measured response time. If the error is less than the allowable range, the performance model is valid. Otherwise the performance model need be modified until the error is less than the predefined real number.