Multi-server based distributed virtual environment (MSDVE) systems have become prevalent, supporting a large number of Internet users. In MSDVEs, the load balancing among servers is an important issue to achieve the system scalability. However, existing approaches must pay high migration overhead, for the state transition of users or regions, thus the excessive holding time during load distribution makes it difficult for the system to keep the interactive performance acceptable.
This dissertation aims to provide an efficient load distribution mechanism in which a group of servers takes charge of regions and shares region information among servers. The proposed mechanism dynamically classifies task types based on the features of requested messages and distributes each task fairly to neighboring servers. We have implemented the proposed mechanism extending our network framework for DVE, ATLAS. Our experiments show that the task distribution reduces both communication and processing overhead during load distribution without significant classification overhead.