الفهرس يوجد فقط 14 صفحة متاحة للعرض العام
 |  | from | 119 |  |  |
| | | from | 119 | | |
المستخلص
Distributed virtual environment (DVE) enables multiple users to interact with each other over a network. Due to the availability of high bandwidth and fast graphic cards, these systems grow in tenns of number of users, scene complexity and interactivity. However, how efficient the system scales, as the number of users increase is the major problem that DVE faced since their inception. Two Key issues for scalability in DVE systems are: load balancing of the overall system workload and a need for workload model to study, analyze, design and tune these systems.
In this thesis, a novel load balancing model for DYE systems has been proposed. The model employs a novel load balancing algorithm which enjoys features missing in existing load balancing techniques. Also, a statistical workload model that characterizes DYE systems has been developed using simulation, in order to study this novel load balancing algorithm. Both models have been validated and verified.
TQe workload model was developed by building a Virtual Environment (VB) prototype of distributed walkthrough application. This prototype is used to generate 180 datasets with different navigation patterns. User requests and inter-arrival periods have been measured. The conducted experiments showed that the developed model is both descriptive and structural in the sense that the distribution parameters are often correlated with different movement patterns, fetching techniques, and objects density of the navigated scene.
1’qe proposed load balancing model is simulated and tested with the proposed workload model. It’ is shown th’tt the proposed load balancing model overcomes the sudden effects of blindly transferring the whole excess load to the least utilized server (i.e. there is only one candida~ sever). Tpis is achieved by distributing the excess load among the rest of system’s servers with high respect to the inter-server communication; by carefully selectin¥ candi@te destmation servers. In addition, it is demonstrated that it outperfonns the ~xistlng globall~d balancing models.
Also, the statistical wortdoad model is used to show, experimentally, that the system saturatiqn point starts earlier than its corresponding point with simple workload for the same n”,mber of clie~ts IUld servers. This demonstrates that the simple workload used in testing e~isting pYE ~Y!ltems are not recommended for forecasting system perfonnance.