Research And Implementation Of Parallel RTI Based On High Performance Computing Environment

The High Level Architecture (HLA) has been widely applied in the field ofmilitary simulation. Along with the improvements of both simulation scale andsimulation resolution, the demand for communication and computing resourcescontinuously increased. As a result, the HLA simulation based on PC network cannotmeet the demand. The high performance computing environment can provide powerfulcomputing and communication capabilities for large-scale simulation applications.However most of current RTI software was based on distributed network environments.The advantages of the high performance computing environment cannot be fullyexploited on the communication structure, message processing, data consistencymaintenance aspects. Therefore, the research on the techniques of HLA/RTI based onhigh performance computing environments, give full play to the platform for efficientcommunication and computing power to improve the operating efficiency of thesimulation system, promote the development of large-scale simulation applications andso has a very important significance.Considering the characteristics of the high performance computing environmentand HLA simulation system, this dissertation aims to improve the performance oflarge-scale or complex simulation system from various factors which affect theperformance of HLA simulation, including the RTI communication structure, messageprocessing, data consistency maintenance methods. The innovations of this paper are asfollows:Firstly, the high performance computing environment tend to provide a variety ofdifferent levels of shared memory/IB/Ethernet hardware communication environment,however, most of the current RTI only provides TCP/IP communications support, it isdifficult to take full advantage of high performance communication of HPC platform.To this end, the paper presents a kind of RTI communication structure supportmulti-level parallel communication, the communication structure using multi-threadedparallel processing technology, communication of data in different hardware assigned toa different thread processing, which can effectively support multi-level parallelcommunication, improve the communication efficiency and throughput of the system.Secondly, the processing of the message in RTI can be divided into two stages:receive unpacking and sort processing. The current RTI often use the single-threadedmessage processing method, two stages can only work serially and the processingefficiency is low. To this end, the paper presents a multi-threaded message bufferqueue-based pipeline processing method, the method adding message buffer queuesbetween two message processing stage, the data is pushed into the message buffer queueafter receiving unpack, and then processed by a special message classifying processing threads. The test results show that: in the frequent message interactions HLA simulation,using the method message receiving and processing efficiency can be increased morethan8%, as compared with single-threaded processing method.Thirdly, data consistency is the premise of the HLA simulation running correctly.The current RTI either use a centralizing strategy, maintenance algorithm is simple, butthe central node often become the bottleneck of the system performance; either usedistributing strategy to avoid system performance bottlenecks, but lead to less efficientand algorithm complexity. Integrating the advantages and disadvantages of the twostrategies, the paper proposed a centralizing and distributing data consistencymaintenance methods, according to the different data modification mode, the Federateand Federation data is divided into two categories: the data modified by federateimmediately(self-modifying data) and the data modified by multiple federatestogether(consultation modify data), the former maintained the autonomy of each localRTI component (LRC); For the latter using centralized maintenance strategy,centralized maintenance by the center RTI components (CRC). This method canguarantee correctness, and improve the data synchronization performance effectively.Fourthly, on the basis of the research results, The shmRTI software on highperformance computing environment is realized which conforms to IEEE1516specification. The test results show that: The shmRTI complete IEEE1516specificationmanagement service applications. In the multi-core PC environment, compared topRTI1516, its intra-node data throughput is more than2.8times, intra-node datacommunication delay can reduces by40%or more; in the network environment mixedIB/Ethernet high-performance computing platform and PC, the integrated time-steppedperformance of shmRTI can increases by more than30%.