SGER: Integrating Actor Model with Real-Time Elastic Control Theory

随着网络技术的快速发展和嵌入式设备的日益普及，传统计算系统的范围已经扩展到对物理环境的智能控制。这种嵌入式混合控制系统存在许多挑战。传统上，编程语言社区关注的是分布式代理之间交互语义的建模和推理，而实时计算社区关注的是如何管理CPU和网络通信资源，以便实时任务可以预测地满足其端到端时间约束。该方法将Actor理论与实时弹性控制理论相结合。实时弹性控制理论是一种将反馈控制器设计与实时调度设计相结合的创新方法。传统上，反馈控制是硬实时应用的典型例子。这项工作允许控制器通过减慢采样频率和调整增益来适应不可预测的工作负载激增。如果没有控制器适配，这种工作负载激增是不可调度的。“弹性”一词指的是控制器期限的动态变化，以改善可用计算资源的管理。这项工作打破了实时调度理论和反馈控制理论之间的界限。将该理论与Actor模型相结合，将打破编程语言与并发控制理论、实时调度理论、反馈控制理论之间的壁垒。除了目前分布式系统的形式化所处理的通常的定性属性（如偶然性）之外，这项工作将为一个统一的框架奠定基础，该框架将允许用户推理定量属性，包括是否可以满足时间要求以及控制下的物理系统是否稳定。

The rapid advancement of networking technologies and increasing use of embedded devices has extended the scope of traditional computational systems to include intelligent control of physical environments. There are many challenges in such embedded hybrid control systems. Traditionally, the programming language community focused on modeling and reasoning about the semantics of interactions between distributed agents, while the real-time computing community focused on how to manage CPU and network communication resources so that real-time tasks can predictably meet their end-to-end timing constraints. This proposed effort integrates the Actor theory with the theory of real-time elastic control. Real time elastic control theory is an innovative approach that integrates the design of a feedback controller with the design of a real-time scheduler. Traditionally, feedback control is a prototypical example of hard real-time applications. This work allows the controller to adapt to unpredictable surges in workload by slowing down its sampling frequencies and adjusting its gains. Such workload surges are unschedulable without the controller adaptation. The word "elastic" refers to the dynamic changes of controller deadlines to improve the management of available computing resources. This work broke the barrier between real-time scheduling theory and feedback control theory. By integrating this theory with Actor model, the barriers between programming language and concurrency control theory, real-time scheduling theory, and feedback control theory will be broken. In addition to the usual qualitative properties (such as eventuality) handled by current formalisms for distributed systems, this work will lay the foundation for a unified framework that will allow users to reason about quantitative properties including whether the timing requirements can be met and the physical system under control is stable.