Research On Software Architecture Performance Based On Queueing Theory

Software quality is always an important issue in the domain of software and most software fail with quality problem. For the software developed is more and more complex, the architecture-based software development mode is the main one; therefore the analysis and evaluation of architecture are necessary at present.Architecture design and analysis depend mainly on experience of architect but not on the theory for a long time. When the software architecture is analyzed, the following questions can be considered: how to measure architecture quality, how to build quality model and what method used to analysis architecture quality. Based on the common software metric theory, the quality attributes are analyzed and the performance model of architecture is put forward in the thesis. Combined with the problem ofperformance model and the result of analysis of design and analysis method, a new method and model of performance analysis are given. Performance is one of architecture quality attributes and considers two aspects of system responsiveness: either the time required to respond to specific events, or the number of events processed in a given time interval.The architecture is designed to solve some problems. Many software architectures may be mapped from the same requirement; therefore some elements of these architectures may be the same so that by comparing these architectures the best architecture can be chosen. The thesis present the method that first we extract the process view from architecture description and then use queueing theory to analysis the process view to build architecture performance model. The process view is one of general architecture views, which describes architectural behavior. Component and data stream constitute a view of process. From concurrent view of process, processes in the view are like nodes in network. We analyze the likeness between nodes in network andcomponents in architecture in process view, between topological structure of network and software architecture, and between messages transmitted by network and data elements processed by software. These likenesses are the basis to analyze the architecture performance with queueing theory that has some good conclusion to meet the analysis of architecture performance. We also debate the maximum throughput in average and average latency of software architecture. By transforming architecture elements (component and connector), we can get the process view of architecture. In the end we use the method to analyze the two cases in KWIC system and compare the results of architecture performance analysis.