Theories and Tools for Sustainable Programming

根据国际能源署2013年的一项研究，电子设备的能耗正以每年6%的速度攀升，是全球总能耗增幅的两倍。加拿大自然资源公司估计，10%的家庭用电进入了电子设备。除此之外，代码重用的有效性导致了代码的大小和复杂性，这使得程序比理想情况下需要更多的硬件和维护资源，并使可靠性的硬保证变得困难。鉴于我们在通信、交通、医疗等方面对计算设备的日益依赖，这些相互矛盾的趋势很难持续下去。我们打算推进编程技术，通过精益编程解决功耗、可靠性和代码重用问题。 在计算的“低端”，传感器网络已经获得了牵引力，例如用于室内/室外指导、城市规划、结构完整性监测、建筑自动化、环境监测、农业和资产管理。这类传感器的超低功耗处理器可以通过收集能量(例如太阳能、热场、运动场、声场、湿场和射频场)供电，预计可以在无需维护的情况下运行多年，例如嵌入混凝土中。我们打算提出一种在能源不可靠的情况下可靠间歇计算的方法。最近，我们开发了一个用于嵌入式系统设计的工具pState，它包括正确性分析、定量分析(如功耗、传输可靠性、消息延迟)、代码生成和最坏情况下的执行时间分析。我们计划通过生成组合使用易失性和非易失性内存的代码来实现间歇性计算，并将pState的分析扩展到间歇性计算的可靠性。 在计算的“高端”领域，自2005年以来处理器核心数量的增加使性能得以提高，同时保持功耗不变。尽管多核处理器在智能手机和数据中心中很常见，但流行的编程模型不能很好地随核数量而扩展。受保护的原子操作很有吸引力，因为它们是“更高级别的”，而且比通常发现的显式信号和消息传递更容易使用。我们最近的工作表明，并发对象中受保护的原子操作可以高效地实现，并且可以随核的数量而很好地扩展。我们计划开发一种基于守卫原子动作的产品级编程语言，并解决三个研究问题：(1)如何正确处理原子动作中的异常；(2)如何通过动态混合来支持代码重用，作为一种灵活而安全的组合机制，与原子动作一起；(3)如何在这种语言中引入所有权，以允许更好的静态检查，简化正确性条件，并提高代码效率。

According to a study by the International Energy Agency in 2013, the power consumption of electronic devices is climbing 6% per year, twice the increase in overall global consumption. Natural Resources Canada estimates that 10% of household electricity goes into electronic devices. Independent of that, the effectiveness of code reuse has lead to code size and complexity that makes programs need more hardware and maintenance resources than ideally and makes hard guarantees about reliability difficult. Given our increased reliance on computing devices in communication, transportation, health, etc., these contradictory trends are difficult to sustain. We intend to advance programming techniques that address power consumption, reliability, and code reuse through lean programming. At the "low end" of computing, sensor networks have gained traction, e.g. for indoor/outdoor guidance, urban planning, structural integrity monitoring, building automation, environmental monitoring, farming, and asset management. Ultra-low-power processors of such sensors can be powered by harvesting energy, e.g. solar, thermal, motion, sound, moisture, and RF fields, and are expected to perform without maintenance for years, e.g. embedded in concrete. We intend to propose a method for reliable intermittent computation with unreliable energy sources. Recently, we developed pState, a tool for the design of embedded systems that includes correctness analysis, quantitative analysis (e.g. power consumption, transmission reliability, message delay), code generation, and worst-case execution time analysis. We plan to implement intermittent computations by generating code that uses a combination of volatile and non-volatile memory and extend pState's analysis to the reliability of intermittent computations. At the "high end" of computing, the increase in the number of processor cores since 2005 allowed an increase in performance while keeping power consumption constant. Although multi-core processors are common in smartphones and data centers, popular programming models do not scale well with the number of cores. Guarded atomic actions are appealing as they are "higher level" and easier to use than explicit signaling and message passing as commonly found. Our recent work has shown how guarded atomic actions in concurrent objects can be implemented highly efficiently and scale well with the number of cores. We plan to develop a production-quality programming language based on guarded atomic actions and address three research questions: (1) how to handle properly exceptions in atomic actions, (2) how to support code reuse through dynamic mixins, as a flexible and safe composition mechanism, with atomic actions, and (3) how to introduce ownership in such a language to allow better static checking, simplify correctness conditions, and improve code efficiency.