The Research Of Information Transmission Latency In Large Scale Distributed Simulations

Owing to the boosting of relevant techniques and current requirements, distributed simulation is enlarging its application field in nearly all dimensions. As a result, more and more simulation entities and LANs are engaged. However, the key to successfully construct such simulations lies in whether the supporting infrastructure can deliver data to participants in a timely manner.In order to improve the simulation system's real-time perceiving performance, this dissertation studies the key techniques to reduce infomiation transmission latency. First of all, the concept of overtime degree is introduced. Through the thorough analysis of the information transfer process, three approaches are presented to cut the data delivery time: properly distributing the entities onto different hosts at the design/deployment phase, dynamically adjusting the entity distribution to fit the system's run-time patterns, and suitably allocating data transmission methods for different information flows.In order to get a good "entity-host" mapping, the criterion is analyzed. Then the problem is transformed into the determination of "simulation space - host" mappings for simplicity, and its NPC hard is also proven. Two heuristic space partitioning algorithms are proposed: Max Flow Merging and Max Possible Benefit Merging. Compared with similar studies, the Max Possible Benefit Merging can produce good results faster.To move entities effectively at run-time, the decision-making system framework is proposed, including a sliding window based simulation monitoring mechanism, a virtual benefit oriented destination LAN selection algorithm, and a conflict-free migration request arbitration algorithm. Guidelines are given on the choice of relevant parameters.As High Level Architecture (HLA) has become the mainstream of the distributed simulation society, the techniques to implement entity migration under HLA/RTI environment are studied in detail. After analyzing the flaws of previous studies, this dissertation presents a novel protocol for moving entities, which fits common understandings well. It has the benefit of not losing messages during migration. It is also proved that TSO messages will not be processed out of order without intention. And the use of the protocol is illustrated.To transfer information more effectively, the Hybrid IP Multicast is originally proposed, which can take advantage of the simulation's communication patterns and physical environment. After analyzing different data transmission methods, this dissertation proposes the Flow Merge with Hybrid IP Multicast First (FM/HMF) data transmission allocationalgorithm, which outperforms its counterparts in experiments.Finally, a simulator is developed to simulate large scale distributed simulations. It is designed to reflect various components included in distributed simulations. Researchers can easily configure the simulator to simulate different distributed simulation systems to carry on various studies.