Research On Key Techniques Of Resource Management System Of Simulation Grid

As a new network computing platform, grid aims to provide an infrastructure for users to make use of the across-the-board sharing resources. The grid technology offers means and supporting technologies to construct this infrastructure. With the advantage of grid technologies, research and constructing of simulation grid (SG) is one efficient way to improve current distributed simulation system, which can particularly overcome some disadvantages in resource allocation, system flexibility, cooperation, security, administrability, and so on.Resources are main objects in SG. How to manage all kinds of resources and provide resource consumers identical user interface are key to construct SG. Therefore, resource management system is a critical component of SG. In this dissertation, we systemically studied the key techniques of simulation grid resource management system (SG-RMS).We firstly analyzed the characteristics of sharing resources in SG, and studied the simulation grid resource (SGR). Afterward, we built the concept model for SG-RMS, and analyzed its function. As a result the centralized-distributed structure model for SG-RMS has been put forward. Based on the model, we designed the framework of SG-RMS, and analyzed the subsystems in the framework in detail. We also discussed the key issues about SG-RMS, including resource organizing, resource co-allocating, and application task scheduling.According to the characteristics of SGR, we presented a formal description for SGR. On the basis of the description, a hierarchical organizing model has been proposed. In this model the SGR resource information are organized by physical resource level and logical resource level. We utilized Light Directory Access Protocol to describe the meta-data of SGR and design SG resource organizing framework. Then, the communication protocol of resource level and organizing methods for SSG (Sub Simulation Grid) have been discussed. In organizing framework, we described resource requirement of SGR, analyzed resource matching algorithms, and resource discovery schemes as well. Resource discovery is classified as inter-domain discovery and intra-domain discovery. We specially studied inter-domain discovery schemes and the mechanism of timeout in resource discovery.In the study of resource co-allocation methods, we mainly resolved the deadlock and livelock issues. We firstly analyzed the reason of deadlock and livelock in resource co-allocation and built resource co-allocation model (RCAM) based on Finite State Machine (FSM). On the basis of RCAM, we established the policies of resource co-allocation. According to single and multiple goal resource allocation states, we designed corresponding resource co-allocation algorithm including SG_RCA, SG_DPRCA, SG_ATRCA, MG_DPRCA, and MG_ATRCA. In order to verify the algorithms performance, we designed and realized the simulation grid resource co-allocation simulator (GRCS) on the basis of the characteristics of SG resource co-allocation. Through simulation test we analyzed the performance and applicability of all algorithms, and proved the correctness of theoretical analysis.In the study of application task scheduling issues of SG, we firstly analyzed the communication mode of distributed simulation application. Through the analysis, we constructed the application task requirement model and resource topology model based on graph theory. With the description of the application task scheduling issues, we defined performance function and correlative performance parameters for scheduling. According to different situations, scheduling issues about single task single node (STSN) and multiple tasks single node (MTSN) were studied. In STSN scheduling issue, we proposed and analyzed heuristic scheduling algorithm named as STSN-TS. In MTSN scheduling issue, we firstly classified the application tasks by means of Hierarchical Agglomerative Clustering (HAC) based on Furthest-Neighbor (FN) method. On the basis of tasks clusters, we designed and analyzed MTSN scheduling algorithm named as MTSN-TS.On the basis of all above research, we designed and implemented the simulation grid resource management prototype system (SG-RMPS). We designed SG-RMPS framework and analyzed all components in the framework. Further, we implemented the experimental system based on Sun Grid Engine (SGE) and Globus components. By running the simulation application which is designed according to XX simulation confront system, we validated the research in this dissertation and the technology feasibility of all research.