Test Prioritization and Localization at Ericsson

爱立信为全球移动的网络开发和提供设备、软件和服务。爱立信在 渥太华开发了蜂窝基站，它是手机与物理骨干网之间的接口 网络.这种高可靠性的环境需要在昂贵的硬件上进行测试， 与基站交互的电话。有些测试运行几天是昂贵的，而另一些只需要几秒钟。 当测试失败时，软件集成过程停止，工程师必须介入。我们的目标是联合收割机 使用经验软件工程进行测试优先级排序，以挖掘包括过去代码在内的历史工件 变更、错误、同行评审和测试运行，以创建统计模型，（1）确定测试的优先级，（2）帮助 开发人员对故障原因进行分类和定位。 对测试优先级排序工作的研究一直受到有关过去运行的信息以及与其他测试的链接的限制。 开发工件。Ericsson收集了大量的日志、测试运行和开发工件的历史， 提供软件的可再现性和审计。当前的大部分分析都是手动完成的， 开发人员必须仔细检查这些信息以找到故障的根本原因。 我的研究团队在挖掘和链接软件工件方面拥有丰富的经验。结合 爱立信丰富的数据存储，我们计划创建统计模型和分类器，以有效地排序测试， 关联其他工件，以提供有关故障根本原因和 导致失败。例如，我们可能会发现，一组短期运行的测试可以找到所有测试的70%以上。 故障，并且由没有经验的开发人员对控制信令的模块进行更改将导致 高冲击失效的可能性增加10%。这些成果将提高生产力， 爱立信渥太华的1000名员工和先进的国家的最先进的了解哪些历史措施 帮助测试优先级和本地化。

Ericsson develops and supplies equipment, software, and services for global mobile networks. Ericsson in Ottawa develops cellular base stations that are the interface between cell phones to the physical backbone networks. This high reliability environment requires testing on expensive hardware that can simulate cell phones interacting with base stations. Some tests are expensive running for days, while others take seconds. When a test fails, the software integration process stops and an engineer must intervene. Our goal is combine test prioritization with empirical software engineering to mine the historical artifacts including past code changes, bugs, peer reviews, and test runs to create statistical models to (1) prioritize tests and (2) help developers categorize and locate the cause of a failure. Research into test prioritization work has been limited by information about past runs and links to other development artifacts. Ericsson collects massive histories of logs, test runs, and development artifacts, in part to provide for reproducibility of software and auditing. Much of the current analysis is done manually where developers must sift through this information to locate the root cause of a failure. My research team has extensive experience in mining and linking software artifacts. Combined with Ericsson's rich data store, we plan to create statistical models and classifiers to efficiently order tests, and to relate other artifacts to provide additional information about the root cause of failures and the factors that contribute to failures. For example, we may find that a certain set of short running tests find over 70% of all failures and that a change to a module that controls signaling by an inexperienced developer will lead to an 10% increase in the likelihood of a high impact failure. These outcomes would increase the productivity of Ericsson Ottawa's 1000 employees and advance the state-of-the-art in understanding which historical measures help in test prioritization and localization.