A Study On Several Issues Of Task Management In Computational Grids

A computational grid is an emerging computing infrastructure that enables effective coordinate access to various distributed computing resources in order to serve the needs of VOs (Virtual Organizations). It is significant and valuable for both researchers and engineers to investigate grid-computing technologies for engineering and scientific computation to support collaboration across different administrative domains. However, grid performance that can be delivered varies dynamically due to resources competing, task uncertainty, and so on. Therefore, resource scheduling and task management are key issues. In grid environments, from task submission to result processing, all events about tasks are under the control of task management. Because grid environments are large-scale, heterogeneous, dynamic, distributed and autonomous, grid task management is complex and challenging. In this dissertation, based on the application background of engineering and scientific computation applications, we focus on several issues of task management in computational grids. All the related reach activities are supported by the National Natural Science Foundation of China under Grant No.60225009 and the National High-Tech. R&D 863 Program of China under Grant No. 2002AA414070.The proposed system of task management is a part of the MASSIVE ( Multidisciplinary ApplicationS-oriented Simulation and Visualization Environment) project, which aims to use Grid technology to establish an enabling environment for distributed simulation and visualization of large-scale scientific and engineering research. In this task management system, there are eight functional modules, such as Job Creator, Job Scheduler, Job Monitor, and Job predictor. The system is of easy usage and good extensibility. Among these modules, Task scheduling, uncertainty handling & QoS (Quality of Service) steering, Task authorization are main topics of our research.In the research work, a distributed schedule model is presented, and is named by D3SM (a Dual-Component and Dual-Queue Distributed Schedule Model). Two relevant scheduling strategies are utilized to improve grid performance: the Task and Data Co-scheduling Strategy adopts a scheduling manner of the collaboration of task scheduling and data scheduling; and the Overlap Strategy of Computing and Data Transferring makes full use of computing resources and networking resources by the overlap technology. An extended Communication-Inclusion Generational Scheduling (XCIGS) algorithm is proposed to schedule dependent tasks of an application with their DAG (Directed Acyclic Graph). During the scheduling, ineligible tasks are momentarily ignored, and a Buffer Set of Independent tasks (BSI) is conducted to leverage the utilization of grid resources. Aiming to the wireless grid environments, a Power-Aware Hierarchical Scheduling Algorithm with respect to resource intermittence is applied in order to minimize the energy consumed and reduce the affection of resource intermittence. To cope with dynamic load balancing and fault tolerance, intelligent agent and redundancy technology are put into force based on the extended D3SM architecture.During the study of grid certainty handling, the D3SM model is extended again by including a few functional components. The scheduling strategy of the Global Scheduler is revised for networking certainty. To deal with task certainty, the order scheduling is adjusted on the base of its strategy adaptability. Three measures are enforced to alleviate effects of resource certainty, such as controlling the length of the STQ (scheduled task queue). To effectively manage grid QoS, an Aggregate Utility Ratio is modeled as a composite QoS, two scheduling methods and steering-enabled visual interfaces are presented. Whilst four performance metrics and aggregate utility ratio are visualized to facilitate the user's interaction with the scheduling, corresponding post-scheduling mechanisms are designed to cope with scenarios where scheduled tasks could not obtain expected QoS.Task Authorization is a critical cha...