Research And Application Of Grid Scheduling Algorithm Based On Cost

As the development of the computer industry (including computer hardware, software and network), the grid concept and technology of early stage have more practical significance. Grid manage and use the idle network resources effectively and make a number of free or advanced equipment in the network easily available to grid users, while a grid scheduling system is the most important foundation for the grid to achieve these functions.The purpose of grid task scheduling is to map the task with a strategy to grid resources to run. Grid users will move to different tasks demand different QoS, while the heterogeneous nature of grid resources, dynamic, autonomous, and distribution and other characteristics, makes the effective task scheduling more challenging under grid environment. A superior or inferior grid task scheduling algorithm is directly related to the implementation of grid tasks and grid system operation conditions, and plays a vital role in the study of grid technology.This thesis studied and analyzed the real grid system, hardware environment and software environment, focusing on analysis of the actual standard Globus grid protocol hierarchy, and the grid development toolkit Globus Toolkit, and uses the environment as the Globus grid computing research and development platform for grid task scheduling in-depth analysis. Analyzes the characteristics of grid resources, the grid scheduler scheduling process and the characteristics of the grid scheduling, it was found that the grid has different needs during the operation of each stage, each step of the scheduling operations affects the final result, the final grid scheduling goals and standards for evaluation of its merits was be explored via above analysis. In addition, the organizational structure of grid scheduling is also an important problem for the effectiveness of the grid scheduling. After a careful analysis of several existing scheduling characteristics of the organizational structure, the thesis proposed the scheduling strategy structure based on fault-tolerant which integrated all the advantages.This thesis in-depth analyzed, compared several classic advantages and disadvantages of grid scheduling algorithms, but also the improvement of classical scheduling algorithms, and extensive research a number of the latest on the task scheduling strategy and ideas, and finally proposed user QoS demand-oriented scheduling system based on index system. The user's QoS requirements will be divided into three categories of high medium low by scheduling system, use BQS (Best QoS Scheduling), ACCOS (Anticipate Compute Capacity Oriented Scheduling), COS (Cost Oriented Scheduling) of three algorithms separately. The cornerstone of three algorithms is the indicator system, the indicator system includes: CC (current computing power), AC (expected availability of computing power), Hot (resource heat), LB (loading balance), LBChange (load change factor), OC (original cost of resources), CT (credit of resource), ACWithCT (computing power and credibility comprehensive index), Cost (comprehensive cost), EC (cost-effective indicators). The indicators Cost and EC are calculated by other indicators, therefore, other Cost Index to check the schedule result by influencing Cost indicator, while the purpose of the establishment of these indicators should also check the grid users and grid resource providers, and grid users and grid resource provider's behavior will affect the corresponding indicators, so as to achieve a healthy interactive system. Three algorithms have different characteristics, to meet different QoS demands, the core idea of BQS algorithm is to provide the best resources for the current task, the best indicator for the primary consideration is CC, CT, and LB; the core idea of ACCOS algorithm is to provide better resources , but the cost should be as low as possible, so all indicators in the index system will be used except CC; the core idea of COS algorithm is to provide the lowest possible cost resources, but capacity should be better as possible.To test the effective of the grid task scheduling algorithm and the role of each index of control, it need different resources, tasks environment, and the methods for test must be easy to experiment, to create various tasks, simulate different resources environment, and the whole environment should can be repeated and controlled, while in the real grid environment, it is difficult to meet such demand, so this thesis used the more popular GridSim grid simulator, and analyzed the simulation techniques and analysis methods of GridSim to realize task scheduling algorithm. Base on above analysis and for the demand of the algorithm and the index system, GridSim was add database access to improve , and in order to facilitate accurate analysis of experimental data, add Excel table output component improvements. In the improved GridSim, the three kinds of grid task scheduling algorithm have been realized in this thesis, and tested the performance and compared.At the end, the thesis effectively realize the practical application of three grid task scheduling algorithm in the real campus computational grid UCGrid3.0 platform, the platform's core technology is based on GT4 development of distributed computing grid middleware, the new post-implementation Grid task scheduling algorithm would be an important component of the platform, significantly enhance the capabilities and availability of grid computing platform for scheduling.