Dissertation > Industrial Technology > Automation technology,computer technology > Computing technology,computer technology > Computer applications > Information processing (information processing) > Computer simulation

The Research of Critical Technique of Resource Management in Virtual Grid Environment

Author LinJianNing
Tutor WuHuiZhong
School Nanjing University of Technology and Engineering
Course Computer Applications
Keywords Simulation Grid SBA Resource Management Task Scheduling Trust Evaluation Data Integration Qos Partial Relation
CLC TP391.9
Type PhD thesis
Year 2006
Downloads 501
Quotes 1
Download Dissertation

Simulation Grid combines grid computing and modern simulation techniques inorder to adapt to the latest characters of model and simulation that are digital, virtual,intemetal, agental and interoperation. Simulation grid will make great progress in thedevelopment in SBA. And the core in simulation grid is resource manage system. Thus,take the project supported by the National Defense pre-research Foundation asbackground, this dissertation mainly develops in resource management system, and makedeep research in the model of resource management, task scheduling algorithms and thetrust evaluation model on the resource nodes.In this paper the related works on the grid computing and simulation model isintroduced as a guide. After analyzing the research on the supporting environment onSBA and the characters of the resources in the supporting environment, the architectureof resource management in the simulation grid is proposed based on the P2P-Grid model,which uses the other resource management system in other grid for reference. In theresource management system, the knowlodges is put into use for the resource descriptionand resource discovery. The trust evaluation system for the resources is also discussed indetail.There are two kinds of tasks in the simulation grid environment. One iscomputation-intensive, the other is data-intensive. In order to meet the real time requestof the application, the scheduling algorithms is put stress on the research for thecomputation-intensive tasks, and three algorithms is addressed. Firstly ascheduling-algorithm based on genetic algorithm (GA) is addressed. The chromosomecoding method and the operator of genetic algorithm are discussed in detail. Therelationship between subtasks can be obtained through the DAG, then the subtasks areranked according to their depth-value, which can avoid the emergence of invalidatechromosomes. In the single resource, two principles are adopted to determine thesequence of the subtasks: one is the short-subtask prior; the other is parent-subtask prior.When the communication delay between tasks is large enough, the task schedulingalgorithms in the grid computing environment based on task duplication is introduced inorder to decrease the communication delay. However, the scheduling algorithms based ontask duplication will generate some redundant task replications, which will enlarge the system loads and even add the makespan. So a new algorithm is proposed to combine theredundant resources and delete the redundant tasks for decreasing the expense of theresource in the grid.Although some iterative methods, such as GA, can solve the problem of taskscheduling effectively. However it will cost too much time when scheduling too manytasks. And some custom heuristic algorithms, such as min-min and max-min, often causethe spare time slots in the resources. So in this paper a new heuristic algorithm isaddressed based on the idea in which the process of scheduling is divided into two steps.First it uses the common heuristic algorithm to schedule the tasks, then a new DAG canbe rebuilt and the decisive tasks and decisive path can be constructed. After that thedecisive tasks will be re-scheduled to the new resource that includes the fit spare timeslots. Also in this paper a new method is adopted to judge the deadlock between tasks.In grid environment grid nodes provide services with QoS guarantee, so how tochoose a credible service is a critical problem. In this paper it is proposed that a kind ofmechanism how to construct basic probability assignment based on the Qos constraintsthat the services provide, and in the process of building the basic probability assignmenta partial relation is inducted in order to retain the worthy history information andconsidering the uncertain information. And referring to behavior of human society thereputation model of the grid nodes is established. Moreover, based on the deferringoperator and the consensus cooperator in the subjective logic theory, the reputationoptions from different grid nodes can be integrated. In order to avoid the overloadappearance, a choosing scheme implemented a biased roulette wheel is adopted.In the meantime, some of the grid nodes that recommend the trust information toother nodes are malice nodes, which often offer the opposite information in order toenhance the node reputation or debase the node reputation. So a new method based ontriangle-fuzzy number is induced. The referring nodes are divided into differentcategories according to their recommendation opinions. And different categories havetheir own deferring operator.Data-intensive application is a kind of important one in the grid environment, whichshould download the files that are distributed in the different grid sites. However, in thepast the scheduling algorithm for the data-intensive application only considered thebandwidth between the source site and the destine node, which omitted the trust level ofthe source site and it is possible to fail in the processing of downloading the files. So anew scheduling algorithm is proposed that considers both the bandwidth and the trust level of the source sites. And an algorithm based on sufferage is addressed to balancingthe download loading in different grid file server site.Simulation grid is a combination of grid and modern simulation and modeltechnology. Many data-intensive applications require a lot of distribute and heterogonousdata resources. Based on the theory researches, in the paper, architecture of the dataintegration system for the distributed and heterogonous data resources is presented basedon the grid, through which the application can fred and bind the data resource at themoment of need. And in the processing of choosing the data resources the model basedon Qos is adopted, which can avoid the fail of data resources effectively.

Related Dissertations
More Dissertations