CSR: SHF: Small: Programming Language, Runtime System, and Architecture Support for Reliability in Intermittent, Energy-Harvesting Computing Devices

新兴的能量收集计算设备（EHD）依靠从其环境中提取的能量运行，有望成为下一代可穿戴设备、植入式医疗设备和物联网（IoT）应用的关键推动者。然而，EHD上的软件只能间歇性地执行，因为能量是可用的。当电源故障（每秒10到100次）时，EHD关闭并重新启动，要求EHD上的软件跨越多个因电源故障而中断的执行周期。不幸的是，今天的软件开发工具和实践并不解决不稳定性，频繁的中断会导致意外和不正确的系统行为。间歇性是EHD广泛应用于许多高价值应用的障碍，并限制EHD仅由专家使用。该提案通过引入基于持久性架构（PCB）的编程和执行模型来解决间歇性的根本挑战。PCB提供了一种新的、更简单的方法来编写软件，这种软件对不稳定性是鲁棒的。PCB还描述了一种新的方法来执行该软件具有高可靠性，尽管不情愿。通过简化为EHD编写软件的过程并使间歇执行可靠，PCB使所有潜在的系统构建者和开发者都可以访问EHD。正如嵌入式计算机系统的编程和执行模型的成熟催生了围绕新嵌入式应用的行业（例如，量化自我项目和家庭传感），该项目中思想的发展推动了EHD的同样转变。 PCB将使研究人员、个人和商业参与者能够创建新的基于EHD的应用程序，而在过去，这些应用程序只能由专业研究人员进行繁琐的专门构建。这项工作的影响预计将超越研究的技术成果。PCB的主要技术成果是编写和执行软件的新方法。程序是由任务组成的，使用一种新的任务图控制流模型，消除了控制流的不连续性，由于顺应性。PCB新颖的基于通道的内存模型与通道访问控制一起确保内存保持一致。PCB简化程序？的推理进程，因为它通过任务图持续跟踪执行，通道访问控制防止任务修改其输入，确保任务重复执行完成时的正确性。PCB跨越系统堆栈的各个层，使用一种使程序员能够使用简单表达式的语言提供语言、编译器、运行时和架构支持。编译器将程序映射到PCB的运行时系统中的操作，该运行时系统反过来实现任务和通道，提供PCB语言所依赖的保证。对通道和任务的硬件和架构支持，通过新颖的硬件支持提供了类似的保证，消除了软件运行时系统的开销。令人鼓舞的初步研究结果支持这样的想法，PCB的任务和渠道，使间歇性计算的推理更简单，PCB消除了正确性问题，在几个实际的EHD应用。

Emerging energy-harvesting computing devices (EHDs) run on energy extracted from their environment and are poised to be the key enabler of next-generation wearables, implantable medical devices, and the Internet of Things (IoT) applications. However, software on an EHD can only execute intermittently, as energy is available. When power fails (10 to 100 times per second) the EHD turns off and reboots, requiring software on EHDs to span multiple periods of execution interrupted by power failures. Unfortunately, today's software development tools and practices do not address intermittence, and frequent interruptions cause unexpected and incorrect system behavior. Intermittence is a barrier to the widespread adoption of EHDs for many high-value applications and limits EHDs to use by experts only.This proposal addresses the fundamental challenges of intermittence by introducing the Persistent- Channel-Based (PCB) programming and execution model. PCB provides a new, simpler way to write software that is robust to intermittence. PCB also describes a new way to execute that software with high reliability, despite intermittence. By simplifying the process of writing software for EHDs and making intermittent execution reliable, PCB makes EHDs accessible to all potential system builders and developers.Just as the maturation of programming and execution models for embedded computer systems spawned industries around new embedded applications (e.g., the quantified-self project and in-home sensing), the development of the ideas in this project propels the same transformation for EHDs. PCB will enable researchers, individuals, and commercial players to create novel EHD-based applications that in the past could only be tediously purpose-built by expert researchers. The impact of this work is expected to transcend the technical outcomes of the research.The primary technical outcomes in PCB are novel approaches to writing and executing software. Programs are composed of tasks using a novel task-graph control-flow model that eliminates control-flow discontinuities due to intermittence. PCB's novel channel-based memory model together with channel access control, ensures that memory remains consistent. PCB simplifies a program?s reasoning progress because it persistently tracks the execution through the task graph and channel access control prevents a task from modifying its inputs, ensuring correctness as the task repeatedly executes to completion. PCB cuts across layers of the system stack, with language, compiler, runtime, and architecture support using a language that enables programmers to use simple expressions. The compiler maps programs to operations in PCB's runtime system which in turn implements tasks and channels, providing the guarantees relied on by the PCB language. Hardware and architecture support for channels and tasks, provides similar guarantees with novel hardware support, eliminating the overhead of the software runtime system. Encouraging preliminary findings support the idea that PCB's tasks and channels makes reasoning about intermittent computation simpler and that PCB eliminates correctness issues in several actual EHD applications.