Research On Time-Space Consistency In Large-Scale Distributed Simulations

Time-space consistency is a basic requirement in distributed interactive simulations. Owing to the boosting of relevant techniques and current requirements, distributed simulation is enlarging its application field in nearly all dimensions. Simulation systems become larger and wider, and more simulation entities are engaged. Consequently, time-space inconsistency comes to be a prominent trouble in large-scale distributed simulations.This thesis addressed some issues related on time-space consistency in High Level Architecture (HLA) simulation.Firstly, the definitions, classifications and causes of time-space inconsistency are presented; a method of quantitative analysis is proposed, including the definitions and calculating methods of various kinds of degree of time-space inconsistency, the designing requirements, as well as three policis to control time-space consistency. The relationship between time advance and requirement of time-space consistency is studied, and as a result, RTI (Run-Time Infrastructure) should play a role as a distributed real-time scheduler and introduce distributed real-time scheduling to overcome the dynamic time-space inconsistency in distributed simulations; the approaches for overcoming various kinds' time-space inconsistency are presented finally.Then, this thesis specifically introduces the new techniques, as follows, for overcoming various time-space inconsistencies.According to the Principle of Locality in large-scale distributed simulation, the conditionality of advancing time among federates are analyzed. And an improved advancing-time strategy, which is based on Region's Prediction, is proposed to avoid current "too conservative" strategy for computing LBTS, Lower Bound on Time Stamp, in HLA/RTI. By extending regions, the strategy can predict the conditionality in future interval. Federate whose time-constrained flag is true can advance time faster, because some federates, without restriction to it, will not be included to compute its LBTS. The optimistic synchronization is used to overcome disorder with little probability. Performance analysis and simulated tests shows the strategy can increase the degree of concurrent, speedup ratio and reduce time-space inconsistency.In order to overcome space inconsistency caused by different coordinates, the coordinate transformation mechanism in HLA/RTI simulation framework is discussed, such as how and when to transform, who accomplishes transformations, and how to designate used coordinate system in messages and to invoke corresponding coordinate transformation module. The modules for transformation between the ordinary coordinates are implemented. The synthetic transformation mode presented, in which the region should be transformed with standard coordinates when a federate declares its interests, while the receivers should transform to its own coordinates when a federate updates its states, can reduce transformations between different coordinates, and cut down the accumulated error in the transformations.According to the characteristics of the delay in WANs, we studied the control mechanism for the probabilistic-weak time-space consistency based on entity migration. Firstly, a criterion topartition entities for time-space consistency is proposed, which tries to make a compromise among load, access locality and communication bandwidth, and the significance entity migration owes to time-space consistency is also presented. Secondly, the advantages and disadvantages that HLA/RTI supports entity migration are studied, especially the entity-migration arbitration policy that satisfied probabilistic-weak time-space consistency. An entity-migration arbitration algorithm based on expected income is proposed. And finally the efficiency of entity-migration arbitration policy is verified by simulated tests.Based on MRMS (More Reward with More Service Time) model we studied the control mechanism for time-space quasi-consistency. At first, the relationship between MRMS model and time-space quasi-consistency in HLA simulations is analyzed. Then a heuristic policy, on-line two-level scheduling policy, is used, in which the top-level algorithm is used to determine the amount of service to allocate to tasks, and the lower-level algorithm is to determine the order to execute tasks by EDF (Earliest Deadline First). By the trade-off between the precision of computing simulated objects and their time-space consistency, consequently time-space quasi-consistency can be satisfied. At last, the simulated tests show that the two-level scheduling policy can improve the time-space consistency of system.Having the development of enabling technologies such as network, real-time operation system and middleware in mind, an extended HLA/RTI framework of time-space strong-consistency is studied. Firstly, the necessity and probability to extend HLA/RTI are analyzed, and the requirement and principle for HLA/RTI extending are presented. Secondly a HLA/RTI architecture with time-space strong-consistency is proposed, including presentation layer, RTI interior-function layer, abstraction layer, operation system layer and communication layer, etc., and the functions and implementation approach of these layers are presented. Finally, the extension in OMT and interface specification is discussed.The current HLA/RTI framework can not overcome time-space inconsistency in large-scale distributed simulations perfectly. So, in this thesis an extension of HLA/RTI based on middleware technique, which can not only encapsulate the standard RTI services but also extend needed ones, is proposed. These extended services can monitor the run-time performance of the simulation system, and be integrated with the time-space consistency technique discussed above. As a result, the extended HLA/RTI framework to overcome the time-space inconsistency can come into being, with the advantages of portability, compatibility and transparency.Finally, based on the research above, a prototype system is developed and tested as well. More improvement could be expected in the future.