Collaborative Research: CSR---EHS: Improving Sensor Network Software Reliability through Language, Tool, and OS Co-Design

无线传感器网络能够从现实世界中收集细粒度的实时信息。传感器网络软件寿命长、规模大、嵌入式程度高，因此必须保证软件的可靠性。该研究项目通过操作系统、其语言和支持程序分析工具的并行协同设计，解决了提高基于组件的无线传感器网络软件可靠性的挑战。该项目专注于TinyOS，并试图解决即使是专业开发人员也会遇到的困难的组件组合问题。长期目标是使非专业开发人员能够从现有组件中创建健壮的应用程序。这项研究基于三种互补的方法。首先，pi正在开发工具支持，为开发人员提供关于如何满足时间限制的建议。在TinyOS中，时序问题很难处理，因为它们以非直观的方式跨越组件边界。其次，pi正在向TinyOS添加对组件接口契约的支持。契约验证使用组件的“规则”是否得到遵守，在开发人员误解或误用接口时指出错误，避免困难的调试会话。最后，基于他们在静态和动态计时以及契约检查之间权衡的经验，pi正在重新审视TinyOS的基本抽象和结构，重新设计它们，使其更容易检查，因此更可靠。其目的是通过呈现设计上不可能出现的许多类错误来提高可靠性，而不是依赖于英勇的分析和测试技术

Wireless sensor networks enable fine-grained, real-time information collection from the real world. Sensor net software must be reliable because it is long-lived, large scale, and deeply embedded. This research project addresses the challenge of improving the reliability of component-based wireless sensor network software through the parallel co-design of an operating system, its language, and supporting program analysis tools. The project focuses on TinyOS and seeks to solve difficult component composition problems that even expert developers encounter. The long-term vision is to make creating robust applications largely from existing components feasible for non-expert developers. The research is based on three complementary approaches. First, the PIs are developing tool support for giving developers advice about how to meet time constraints. Timing problems are difficult to deal with in TinyOS because they cut across component boundaries in non-intuitive ways. Second, the PIs are adding support for component interface contracts to TinyOS. Contracts verify that the "rules" for using a component are respected, pinpointing errors when developers misunderstand or misuse an interface, avoiding difficult debugging sessions. Finally, based on their experiences with the tradeoffs between static and dynamic timing and contract checking, the PIs are revisiting the basic abstractions and structure of TinyOS, redesigning them to be more easily checkable, and therefore more reliable. The intent is to improve reliability by rendering many classes of bugs impossible by design, rather than relying on heroic analysis and testing techniques