|
The rapid development of network makes it possible to form the grid. Utilizing Internet, Grid organizes distributed computers into a virtual supercomputer, and manages idle resources (including computing resources, storage resources, and scientific instruments, etc.) effectively, makes it available to grid users. Accompanying the development of grid technology, computing grid is developing rapidly all over the world.Grid task scheduling is that, how tasks will be mapped to run on the grid resources in order to achieve the best scheduling strategy. However, because grid resources have the following characteristics: Heterogeneity, dynamic, self-government, as well as the distribution, etc., grid task scheduling is a challenging problem. Grid task scheduling algorithm is directly related to the scheduling of the speed, quality, etc., and plays an important role in the grid study. Improving the existing grid scheduling algorithm to meet the objectives of different scheduling as well as possible, is also important.This thesis starts from the grid architecture, studies the protocol of GT4, and studies grid resource characteristics, compares of grid scheduling and local scheduling and distributed scheduling differences, focuses on the structure of the grid scheduling system as well as evaluating the goals, and establishes a grid task scheduling model.This thesis analyzes and compares a number of classic grid scheduling algorithms, in particular, studies of the MIN-MIN algorithm deeply, points out the lack of MIN-MIN algorithm, and proposes several improved algorithm in several ways. This thesis, aiming at MIN-MIN algorithm with the defect of a low utilization rate of resources and a long execution time, considering multiqueue thinking, proposes SEG-MIN-MIN algorithm. SEG-MIN-MIN algorithm divides tasks into several queues by the length of tasks, and schedules the long task queue first, which can save the total execution time. Considering the execution time and execution costs as well as grid users level three factors, this thesis proposes priority-based scheduling algorithm, MMPRI algorithm. Using MMPRI algorithm, under the same conditions, the higher priority that the user owns, the earlier whose tasks will be scheduled. As considering the impact on scheduling of the cost, MMPRI algorithm reaches a balance between total execution time and implementation costs to some extent. Because MIN-MIN algorithm may lead to load imbalance, combining with MAX-MIN algorithm, this thesis proposes a cycle scheduling algorithm, RMM algorithm. RMM algorithm dynamically chooses to use MIN-MIN or MAX-MIN algorithm according to the balance of the current system, can effectively improve the system’s balance.In order to test the effectiveness and performance of grid task scheduling algorithm, it needs to be tested in different situations. Besides what we need is a repeat, and can be controlled environment. As a result of the use of real grid environment to do it is difficult, we use the grid simulator to complete task scheduling algorithms experiments.This thesis carries on a detailed comparison about several commonly-used grid simulators especially the GridSim, and explores the simulation technology and debugging analysis methods of task scheduling algorithms by using GridSim. Using java programming technology, this thesis realizes the MIN-MIN algorithm and the improved algorithms proposed in this thesis on GridSim, compares and analysis of the performance. The result of experiment shows that the algorithms proposed in this thesis are valid.The core technology of ChengDu University of Technology (CDUT) University Computing Grid 3.0 (UCGrid3.0) platform is a distributed computing grid middleware which is developed on the base of GT4. UCGrid3.0 builds campus computing grid portal, and provides the job management, resource management, user management, VO management, plug-in management, node management and so on. This thesis adds resource register function into the UCGrid3.0, improves the campus computing grid platform further.
|