Modeling and prototyping of agent based Resource management and scheduling for Qos oriented grid

Abstract

Increase in the demand for resource requirements and CPU cycles by complex scientific and e-business applications has driven the need for dynamic collaboration of many heterogeneous computing resources to be able to work together. Grid computing involves harnessing the diverse heterogeneous resources to solve compute and resource intensive problems. Compute grids are complex systems composed of remote heterogeneous non-dedicated systems and resources which interoperate to fulfill the resource requirements of compute intensive applications. Such types of distributed systems are inherently unreliable with high chances of different types of failure. Different types of problems that can occur in grid systems are - dynamic nature of resources with resource leaving and joining the grid and change in resource attributes and status, need for high availability and performance to execute compute intensive problems, high probability of failures and unreliability. Resource management and scheduling is the central and important component of grid system which can provide dependable, predictable, sustained and desired level of performance by providing Quality of Service (QoS) assurances to grid end user applications. Extensive literature study related to the research area and survey of related work done to investigate the issues and challenges in QoS oriented grid resource management and scheduling. Based on the findings of literature study, the need and requirement analysis for QoS oriented grid model, QoS algorithm and mechanisms has been proposed. The best-effort approach is inappropriate if grids are to be used as the infrastructure for real world commercial applications and complex scientific applications like simulations and modeling. The QoS oriented grid enhances and leverages the basic job submission functionality provided by grid middlware like Globus and Sun N1 grid engine with QoS support. QoS oriented grid provides a grid resource broker (or meta scheduler) to abstract the users from comp