Cooperating Autonomic Agents for Management of Complex Systems

自主计算是一种通过自我管理或自主管理来降低复杂计算系统总拥有成本的方法。自治管理范式包含两个系统：要管理的系统，由要管理的对象及其属性和关系组成；以及管理系统，管理代理、服务和组件。然而，简化复杂计算系统管理的主要目标仍然面临着实质性的挑战：a)设计和开发自主代理仍然具有挑战性，为一个系统开发的代理不容易适用于另一个系统；B)复杂系统通常由子系统组成，这些子系统也必须使用多个代理进行管理；C)代理可以独立运作，但有时必须合作，这增加了设计和运作的复杂性；D)管理系统变得更加复杂，需要自己的自动化工具。我的研究通过开发新的方法和方法来解决这些挑战，这些方法和方法用于指定要管理的系统的自主代理和领域特定方面。本文将重点介绍用于指定自治代理、要管理的系统、简化多个代理及其交互的规范、支持重用以及用于管理的底层运行时元素的方法的创建。我的愿景是创造新的方法来促进自主方法的设计、使用、操作和应用，以管理大型复杂系统。我将创建一个规范框架，用于定义管理代理、有关被管理系统的相关域信息和用于自动生成代理的方法。该框架将基于一种新的面向对象语言，用于指定代理、领域对象和策略的抽象定义。与被管理系统相关的信息将通过对象类、它们的属性和关系以及特定于领域的管理元素的规范来指定。代理协调的新方法将包括新的基于策略的结构来表达协调策略，以及将协调策略嵌入到利用“高阶”策略的“高阶”自治管理器中的新方法。我将开发新的方法，使我们的自主元素能够体现主动管理，其中自主代理将建立预测模型，预测潜在的问题并采取行动，从而可以避免这些问题。研究结果将直接解决使用自主代理管理大型复杂系统的现有挑战：简化自主代理规范和在多领域重用的创新方法和方法；基于策略的复杂系统多智能体管理新算法与方法支持主动管理的新方法。

Autonomic computing was proposed as a means of reducing the total cost of ownership of complex computing systems by means of self-management, or autonomic management. The paradigm of autonomic management encompasses two systems: the system to be managed, comprised of objects to be managed, their properties and relationships, and the management system, the management agents, services and components. However, the primary objective of simplifying management of complex computing systems still faces substantial challenges: a) designing and developing autonomic agents is still challenging and agents developed for one system are not easily adapted for another; b) complex systems are often comprised of subsystems which must also be managed requiring use of multiple agents; c) agents can operate independently but at times must cooperate, adding to the complexity of the design and operation; d) the management system becomes more complex requiring its own automated tools. My research addresses these challenges through the development of new approaches and methods for specifying autonomic agents and domain specific aspects of the system to be managed. It will focus on the creation of methods for specifying autonomic agents, the system to be managed, simplifying the specification of multiple agents and their interaction, enabling reuse, and underlying runtime elements for management. My vision is to create novel methods to facilitate the design, use, operation and application of autonomic methods for the management of large, complex systems. I will create a specification framework for defining management agents, associated domain information about the managed system and methods for automatically generating agents. The framework will be based on a novel object-oriented language for specifying the abstract definition of agents, domain objects and policies. Information germane to the managed system will be specified through classes of objects, their properties and relationships, as well as specification of domain-specific management elements. New approaches for agent coordination will include novel policy-based constructs to express coordination strategies and new approaches where coordination strategies are embedded in "higher-order" autonomic managers utilizing "higher-order" policies. I will develop new approaches to enable our autonomic elements to embody proactive management where autonomic agents will build predictive models to anticipate potential problems and take action so they can then be avoided. The results of the research will directly address existing challenges in the use of autonomic agents to manage large complex systems: innovative approaches and methods for simplifying the specification of autonomic agents and reuse in multiple domains; novel algorithms and methods for policy-based, multi-agent management of complex systems; new approaches supporting proactive management.