Research On The Technologies Of Message Order Management For Large-scale Distributed Simulation

Currently, the HLA (High Level Architecture) is well-known and used worldwide and time management services are the critical technology in HLA. However, because of the highly-couple relationship between RTI (RunTime Infrastructure) and simulation time, the actual application of time management services met unprecedented challenges, and especially with the expansion simulation there are some disadvantages in distributed simulation application using Timestamp Order (TSO) time management, such as low performance, hardly interoperatility etc. For example, RTI was applied without time management in some large-scale simulation systems, such as "Millennium Challenge 2002 (MC02)" and "The Joint Urban Operations (JUO) experiment", where causal abnormity may be prone to come forth.In the dissertation, aiming at the questions coming from the current time management, based on the hierarchical structure in distributed simulation, causal time is introduced in the infrastructure, the method dividing the causal time and simulation time is proposed, and the message order management protocol is researched. The main contributions of the dissertation are summarized as follows.1. Aimed to advance the usability of the architecture and construct the hierarchical architecture for messages transmitting orderly, the logical time without semantic is introduced to infrastructure to ascertain the causality between messages after the logical time in distributed systems is addressed. Accordingly, the modularizing and hierarchical architecture (Multi-Layer Architecture, MLA named in the dissertation) is put forward to enable the information encapsulation and anti-encapsulation between the different levels. Inside the MLA, the Infrastructure, Message Transport Infrastructure (MTI) named in the dissertation, assures the message causality between federates using causal time while providing service interfaces to federates as RTI middleware. It is in the Federation Layer that simulation time can be used to synchronize federates during the course of running simulation. Because of the method of separating the causal time and simulation time, the autonomy of simulation components can be reached, the concurrence between simulation entities increased, so that the performance in distributed systems will be improved, which helps the application of large-scale simulation. Moreover, it is possible to reuse compenents in the different levels of model, federate and federation.2. In distributed simulation of client-server architecture, aiming at the questions coming from the current time management, the structures of distributed simulation using MTI as server is firstly analyzed. The resolution using First Come First Service (FIFS) queue in MTI is secondly brought forward based on the assumption that the net transmits message reliably and orderly, which can formally prove the conclusion that the causal abnormity can be avoided using the resolution. The message order management protocol can not only assure the causality between messages, but also advance the performace obviously.3. In distributed simulation of peer-to-peer architecture, the distrbuted MTI is applied to aviod the causal anomaly between messages. Firstly, the condition under which the casual anomaly may occur is analyzed in simulation applications based on distributed architecture, which is that direct path and indirect path for transmitting messages between simulation nodes coexist. And then, based on the Einstein-Minkowski relativistic space-time view, the method of vector time management using publish-subscribe topology is put forward, in which each local MTI component detects the received messages and only the effect message firstly received is buffered until its all causal messages have been delivered to the federate. With the method the casual anomaly can be avoided, the concurrent increased availably. In conclusion, the performance in distributed simulation can be enhanced.4. As for the hybrid of client-server and peer-to-peer, the dissertation describes the federate proxy enabled Hierarchical Federation Community (HFC) supporting interoperability and reuse of federation in large-scale distributed simulation. Aiming at the question that causality is difficult to assure in federation community, a kind of hierarchical Time Stamping System for HFC proposed is delivered to ensure the causality between the messages and the message order management regime using divide-and-conquer approach is brought forward, which provides correct operation between remote federations. The message order management regime is not only proven correct, but also found with some merits such as performance improvement, the zero errors, information hiding and supporting interoperability among heterogeneous federations utilizing different internal TM regimes.5. For realizing the effective development in distributed simulation application using MLA, the modeling method based on Agent is described in the dissertation. The application domains and application instances of the MLA and message causality are given at the end of the dissertation.