Agent-based Resource Management And Task Scheduling In A Grid

A computational grid is a hardware and software infrastructure that provides dependable,consistent, pervasive, and inexpensive access to high-end computational capability. Thegoal of grid is to use equally and share dynamic resources among diverse virtualorganizations and coordinate to solve issues. A grid integrates distributed in geography,heterogeneous, dynamic, all kinds of high performance resources, including remotecomputers, network, storage equipment, diversified science instruments, visualizationand virtual reality display equipment and personal computers and so on, to constitute a sopowerful independent computing resource on the user's table, thereby users can use theseresources transparently and seamlessly to solve nearly all kinds of problems and not toneed consider geography location of resources. Resource management is an importantinfrastructural component of a grid computing environment, and its goal is mainly toaccept application requests for resources in a grid and then to allocate resources for anapplication according to an effective scheduling policy in order to ensure task executioncorrectly.The availability and acceptability of a resource management system will rely onscheduling policy. The dynamic and heterogeneous characteristic of resources determinesthe complexity of scheduling systems in resource management systems. Therefore thescheduling policy plays a very important role in performance of a resource managementsystem.Agent techniques have become an important software development technology. Agenttechniques provide a new pattern for software system design to solve cooperate issueamong complex systems. It emphasizes the autonomy and criterion. An agent possessautonomy, sociality, activity and adaptability, which make it adapt to large complexsystems well.Agent methodology is introduced for solving resource management problems in a grid.Each agent is viewed as a representative of a grid resource at a meta-level of resourcemanagement. Therefore, an agent can be considered a service provider. Agents areorganized into a hierarchy with service advertisement and discovery capabilities. Agentsmay discovery flexibly resources for applications and manage dynamically.Multi-Agent hierarchical model solves the scalability and adaptability. The advertisementand discovery functions of resources make it to discovery automatically availableresources after a task request was submitted, which solves the resource scheduling issuein large-scale systems. In the algorithm, the agent will trigger a discovery process to finda resource that can execute the task when an application request is submitted.If the available resource information is found in the upper layer, but the resource may beexecuting other task now and the operation that modifies state information has not yetbeen finished, this discovery results in a feint, that is to say, discovery is successful butthe request cannot be met. In the extreme circumstances, maybe a discovery process findsrepeatedly such resources all the time. Ultimately, the request can never be met.To solve the problem, we bring forward a service advertisement and discovery strategyusing task requests as the object. After a task is submitted, advertise the task in the agenthierarchy, and will first notify agents while some resources, and then to discoveryexecutable tasks to ensure executing the tasks as early as possible. The strategy avoidedthe above-mentioned problem and can attain the goal. Another problem lies in the papersabove is that, the organization of agents is hierarchical structured, as the utilization ofresources changes with time going by, the sub-tree with the current agent as the root willinvalidate if the agent invalidates, that is, arising resource island. And thus result in theinvalidation of resources, which is what we do not want to see. If the agents are organizedin a graph, this problem will be solved. Agent model organization based on graph isintroduced in this paper, and it broadcasts tasks on this structure, heuristic searchingmethod is adopted to find tasks, and thus this problem is solved.This paper brings forward new several kinds of schemes to resolve correspondingproblems with regard to the grid resource management on the basis of analyzing somesolutions based on agent.The main contributions of this work are listed below:(1)The roles and research of the grid resource management are summarized. The thesisintroduces the background, architecture, constitute and the key technologies of gridresource management. Furthermore, as a research foundation, the thesis also analysis anddiscuss farther the research status at present, advantages and deficiencies of agent-basedresource management.(2)We realized the advertisement and discovery requested from tasks of users by theadvertisement and discovery function of messages in multi-agent hierarchical structures,thereby to achieve the management for the grid resources and task scheduling. Themethod solved the message dating problem that is caused by the communication delayduring the broadcasting and discovery for resources in multi-agent hierarchical structures.Thus utilized resource ratio can be increased and at the same time the user task can befulfilled as soon as possible.(3) A multi-agent-based graph structure model is presented. The model solved well theresource island problem, that is, the sub-tree of the agent will become an isolated node ifsome agent lose efficacy.(4) A heuristic algorithm is designed to discovery resources and tasks based onmulti-agent hierarchical structures. The algorithm improves the performance of theresource management and task scheduling systems.At last, the contents of the paper are summarized and the future works to do areproposed.