Assessing and Detecting Components Suffering from Sub-optimal Performance for a Web-based CRM System

性能是关系到软件系统成功和可持续发展的关键因素之一。**如果不能提供令人满意的性能，将导致客户放弃和收入损失。例如，亚马逊报告称，网页加载延迟一秒可能会导致**他们每年的销售收入损失16亿美元。然而，由于现有工具的限制，寻找改进和优化系统**性能的领域是非常具有挑战性的。现有的性能分析工具只能为每个单独的**组件提供资源或计时信息，而不是直接指向次优组件。由于组件在执行实际任务时会消耗处理时间和资源，因此它们**可能不是性能不佳的领域。**需要更复杂和深入的分析**以发现性能改进的潜在领域。**Sparky是一个基于Web的软件程序，负责CRM(客户关系管理)和为Copywell的高速喷墨打印机作业报价**。现有的流程复杂且耗时。提高Sparky的**效率和可伸缩性将大大增加可处理的任务数量，从而**导致Copywell的收入增加。我们通过一个三步流程来评估和改进Sparky的**架构。首先，我们将通过调查领域专家和**挖掘现有的历史使用数据来表征Sparky的工作负载。其次，我们将通过执行**性能测试和构建性能模型来评估Sparky的性能。第三，将应用各种方法(架构级分析、**设计/代码级分析和性能反模式检测)来检测基于结果模型的次优**性能组件。

Performance is one of the crucial factors related to the success and the sustainability of software systems.**Failure to provide satisfactory performance would result in customers' abandonment and loss of revenue. For**example, Amazon reported that one second delay in loading their webpages could result in $1.6 billion loss in**their sales revenue annually. However, it is very challenging to find areas to improve and optimize system**performance due to limitations of existing tools. Instead of directly pin-pointing the sub-optimal components,**existing performance profiling tools can only provide resource or timing information for each individual**component. As components consume processing time and resources while executing the actual tasks, they**might not be the areas suffering from sub-optimal performance. More sophisticated and in-depth analysis is**required to detect potential areas for performance improvement.**Sparky is a web-based software program responsible for CRM (Customer Relation Management) and quoting**for high speed ink-jet printer jobs for Copywell. Existing process is complex and time consuming. Improve the**efficiency and scalability of Sparky would greatly increase the amount of tasks that can be handled, and hence**results in an increase in the Copywell's income. We process a three-step process to evaluate and improve the**architecture of Sparky. First, we will characterize the workload of Sparky by surveying the domain experts and**mining the existing historical usage data. Second, we will assess Sparky's performance by executing**performance tests and building performance models. Third, various approaches (architectural-level analysis,**design/code-level analysis, and performance anti-pattern detection) will be applied to detect sub-optimal**performance components based on the resulting models.