Research On Classifying-Based Grid Scheduling Model GSM-C

As a key branch of DC (Distributed Computing), grid technology has nowadays evolved greatly in scientific HPC (High-Performance Computing) and Internet-based business scenarios. With the increasing requirements for distributed computing, researchers have turned their works from traditional cluster computing to Internet-based and Web service-enabled heterogeneous computing. In this process many great infrastructures, standards, models, algorithms and middle-ware products have been welling up, such as the OGSA(Open Grid Service Architecture), .Globus project, Condor scheduling systems, and GrADs systems, etc.Grid computing can offer the sharing of heterogeneous resources, maximizing the capacity of them. However, grid computing is facing the problems of multi-domain resource controlling, load balancing, quality of service (QoS), and security, just as other distributed computing technologies. Resource discovery and task scheduling are two challenges due to the grid's inherent dynamic feature and are still need to be drilled down into.Currently, domestic and foreign researches on resource and task scheduling are mainly based on two resource organizing models: central model and non-central model. And resource's information is gathered by classifying them as recent information and relative static information. These researches have a common bottle-neck: the long cycle of information retrieval delays the scheduling strategy making and leads to the poor quality of service. In this paper a classifying-based grid scheduling model (GSM-C) is presented. In this model the characteristic of resource capacity is denoted by the class ID which is defined periodically by the classifying strategy with regard to the dynamic features of the resource. This paper is presented according to the following three aspects of research works.Firstly we introduce and analysis recent research on the grid resource management and scheduling.Secondly we design a classifying-based grid scheduling model (GSM-C), in which we design a classifier which generates class ID for each resource and task, and these class IDs can be referenced by the scheduler's mapping module. The task...