Research On Key Technologies Of Application Layer Multicast For Distributed Interactive Simulation

Distributed interactive simulation has been widely used in the many fields of national economic and defence. With the rapid increase of the complexity of the simulated system and the expanding of simulation scale, thousands of state update messages for the entities might be transmitted between numbers of simulation members in the distributed interactive simulation. It has become an urgent issue to provide efficient WAN multicast services with QOS guarantee and reduce the network bandwidth consuming during simulations in the distributed interactive simulation applications.Because of the deployment limitation of the IP multicast, researchers present that application layer multicast (ALM) may be used to provide WAN multicast services. However, the existing ALM protocols are mainly focused on the construction, maintenance and optimization of the overlay networks and delivery trees for applications such as video meeting, IP TV etc. Obviously they would not be well suited for multi-group and multi-source applications such as the large-scale distributed interactive simulations.This thesis expands in-depth research on ALM for distributed interactive simulation aiming at the requirements and characteristics of simulation applications and the deficiencies of the existing ALM protocols. The main contribution of our work can be divided into four aspects as follows:(1) The synthesis classification and evaluation method of the ALM protocolsThe existing classification methods of the ALM protocols are more constricted on the system framework and the constructure of the ALM system, but less on the optimization objects and the optimization technologies used. This paper presents a synthesis classification and evaluation method of the ALM protocols, which synthetically considers most problems in the design of ALM protocols, including five aspects as the data delivery strategies, service models, performance optimization objects and techniques, and fault tolerance mechanism. Furthermore, some recommendations are presented for the design of ALM protocols.(2) The degree and diameter bounded minimum spanning tree algorithmThe self-organization algorithm of end systems is the key function of the ALM protocols. Most researches on the self-organization algorithm are dedicated in the construction, maintenance and optimization of the ordinary overlay networks, and they pay less attention to the specific application background. This paper presents the degree-bounded and diameter-bounded minimum spanning tree problem (BDBDMST), and proposes a centralized arithmetic CST and distributed algorithm DBST. Compared to other ALM protocols, DBST implements the core placement and core migration and is able to accomplish the performance of the multicast tree would not decrease dramatically when the group members join and leave frequently.(3) The degree and diameter bounded multiple shared trees algorithmTo solve the the problems of single delivery tree such as single-point failure, longer delay and higher traffic concentration, this thesis presents a degree and diameter bounded multiple shared trees algorithm, called MDBST. MDBST uses the RTT between nodes as the metric of distance, and implements the even distribution of the multiple cores by maximum the distances between cores. Besides, the node that has the maximum available bandwidth is most likely to be elected as a new core. In order to accommodate the strong dynamic properties of the group members, MDBST supports the core migration according to the state of multiple shared trees. The experiments show that MDBST is maximum-delay bounded, traffic load balance and multi-group supported, and it is well suited for the multi-source and multi-group environments of the distributed interactive simulation applications.(4) The traffic control algorithm based on the priority queue for application-layer multicastThe existing ALM protocols achieve the congestion control relying on the TCP protocol. But this mechanism can not meet the requirement of the large scale environments. In order to adapt to the distributed interactive simulations, this paper designs an ALM traffic control algorithm based on the priority queue, named GFC. GFC is composed of flow-control mechanism and feed-back mechanism. The former mechanism allocates the bandwidth and schedules the delivery of packets based on the priority of the data and the length of the sending queues; the latter mechanism emendates and adjusts the traffic control parameters based on the feed back information of the TCP. The experiment results demonstrate that the GFC algorithm has many good characteristic, such as scalability and stability.According to these studies we design and implement an application layer multicast system named DVSCast for distributed interactive simulation. In DVSCast multicast trees are constructed with algorithm DBST and MDBST. Comparing to XOM, DVSCast has lower control overhead in computing and saving multicast trees and stronger adaptability to dynamics. DVSCast also implements GFC to support end-to-end flow control in overlay networks.In summary, our work presents solutions to several key problems of the ALM in the large distributed interactive simulation, and has academic and practical value for advancing the use of ALM in the distributed interactive simulation.