Distributed Supervisory Control: Theory and Applications

本文从分布式调试和分布式仿真两个方面提出离散事件系统的分布式监控理论。以前在分布式控制中的大部分工作都是基于控制器不通信的假设。没有沟通能力的控制会导致非常限制性的行为。近年来，在分布式控制理论中引入通信的研究取得了一些重要成果。Ricker和Rudie将模态逻辑用于监督控制。巴雷特和拉福恩还解决了沟通问题。它们通过与监管人员沟通（对未来沟通有短视和非短视的期望），为控制提供了必要和充分的条件。虽然这些工作提供了许多见解，但它们在实际系统中的应用有局限性，例如，它们假设没有通信延迟，并且控制器在系统的初始状态上是同步的。这些假设在当今的通信网络中是不正确的，通信延迟在决策中起着重要作用。本项目的目标是消除这些假设，然后将该理论应用于一些分布式应用程序。我们要考虑的第一个应用是调试分布式程序。分布式调试中的许多问题都可以归结为观察和控制的框架。虽然分布式程序的观察问题已经被许多研究者研究过，但是控制问题却很少受到关注。自动控制的抽象为用户测试分布式程序提供了极大的便利。控制消息排序的能力对于程序员使用不同的消息和事件排序来测试程序非常有用。类似地，如果分布式程序与外部观察者交互，则不同进程的相对速度可能会影响结果。在提出的方法中，系统将自动控制各个过程的速度，使期望的谓词变为真。我们要考虑的第二个应用是分布式模拟。分布式仿真主要有保守和乐观两种方法。最近，我们开发了第三种方法，称为k -乐观算法，它允许在程序执行期间改变乐观度。我们的假设是，在适当控制乐观度的情况下，k -乐观算法可以比保守算法和乐观算法运行得更快。我们建议开发一种用于分布式仿真的自适应控制算法，并相对于传统算法测量其性能。我们在这里注意到，提议的项目需要在离散、事件系统和分布式系统的监督控制方面的专业知识。我们在这两个领域都进行了广泛的工作，项目的目标之一就是从一个学科到另一个学科的思想交叉施肥。

We propose to develop theory of distributed supervisory control of discrete events systems in the context of two applications - distributed debugging and distributed simulation. Most of the previous work in distributed control is based on the assumption that the controllers do not communicate. Control without the ability to communicate leads to very restrictive behavior.Recently, there has been some important work to incorporate communication in the theory of distributed control. Ricker and Rudie have used the modal logic for supervisory control. Barrett and LaFortune have also tackled the problem of communication. They provide necessary and sufficient conditions for control by communicating supervisory controllers (with myopic and non-myopic expectations of future communication). Although these works provides many insights, they have limitations in application to real systems, For example they assume that there are no communication delays and that controllers are synchronized on the initial state of the system. These assumptions are not true in the present day communication networks in which communication delays play an important role in decision making. The objective of the proposed project is to remove such assumptions and then apply the theory to some distributed applications.The first application we will consider is that of debugging distributed programs. Many of the problems in distributed debugging can be cast into the framework of observation and Control. While issues in observation of a distributed program has been studied by various researchers, the issue of control has received less attention. The abstraction of automatic control provides great power to the user while testing a distributed program. The ability to control the message ordering is useful for the programmer to test the program with different message and event ordering. Similarly, if a distributed program is interacting with an external observer, the relative speed of different processes may affect the outcome. In the proposed method, the system will automatically control the speed of various processes so that the desired predicate becomes true.The second application that we will consider is that of distributed simulation. There are primarily two approaches for distributed simulation - conservative and optimistic. Recently, we have developed a third approach called K-optimistic algorithm which allows the optimism to be changed during the execution of the program. Our hypothesis is that with the proper control of optimism, the K-optimistic algorithm can run faster than both conservative as well as optimistic algorithm. We propose to develop an adaptive control algorithm for distributed simulation and measure its performance with respect to traditional algorithms.We note here that the proposed project requires expertise in supervisory control of discrete, event systems as well as distributed systems. We have worked extensively in both of these areas and one of the goals of the project is cross-fertilization of ideas from one discipline to the other.