| Summary: | 博士 === 國立中山大學 === 資訊工程研究所 === 87 === This dissertation explores potential methods and practical methodologies for constructing a networked virtual environment (NVE). We first describe the design of distributed and collaborative technologies in which it allows multiple users to interact with each other in a collaborative manner, and then exploit distributed processing technique to speed up the network computing towards the 3-D graphics rendering. In order to evaluate the performance of applications on network computing environment, we have developed an applicable experimental protocol and have applied the protocol to the areas of 3-D graphics rendering. A new load-balancing algorithm with the fault tolerance ability, called global distributed control (GDC), is used to balance the workload on a cluster of network workstations. Experimental results show the GDC performs the best compared with three traditional load-balancing algorithms. To regulate the collaborative behaviors, we present a simplified collaborative group definition (SCGD), collaboration policies, and the finite state of participants to dynamically manage participants within the multi-user environment. Based on our proposed collaborative mechanism, the server can efficiently determine the status of collaboration activities.
To provide the best level of information consistency and support maximum number of participants in an NVE, a three-level control mechanism (3LCM) is proposed to maintain the dynamic shared state for shared objects. The 3LCM consists of three major techniques for dynamic shared state maintenance and can automatically switch from one level to another level based upon some predefined criteria to maintain maximum performance. An approximate Poisson discrete distribution is used to establish the performance evaluation model. Experimental results show that the proposed 3LCM is very effective in maintaining low network latency over a wide range of participants and performs much better than the shared repository technique.
Furthermore, the 3LCM has been implemented to the easy-to-establish HLA-based middleware, federateServer, in order to maintain information consistency and minimize message traffic for distributing information among client hosts and federateServers. Based on the Run-Time Infrastructure (RTI) services within the IEEE standard P1516 - High Level Architecture (HLA) and Java application programmer''s interfaces (APIs) of the RTI, the proposed HLA-based platform provides an architectural foundation to enhance interactivity, portability, and interoperability in an NVE. In addition, the proposed platform design not only provides a client/server capability for simulation on the Web but also supports distributed federation executions over the network.
In terms of the goals of HLA Data Distribution Management (DDM) services, namely efficiency, scalability, and interfaces, a dynamic filtering scheme (DFS) is proposed to reduce "false positive" updates and increase the accuracy of filtering ability to the HLA applications. Due to a reduction in the extension of the regions can decrease false positive updates. This scheme depends upon a well-defined function of filtering efficiency to dynamically change the position of the subscription regions or reduce the abstract volume of regions in order to avoid unreasonable extension of the subscription regions. To verify the feasibility of the proposed HLA-based platform, a distributed discrete-event simulation application in Java is developed and performance comparisons of the proposed modeling design with the distributed object model of Java Remote Method Invocation (RMI) are also presented. In our experiments, we compared the turnaround time by increasing the number of federates within the federation. From the experimental results, we have shown that the design of distributed and collaborative platform based on HLA using the 3LCM is workable and practical for supporting a large-scale NVE.
|
|---|
