Research On Support Technology Of Cloud Simulation For Large Scale Complex System

Large-scale Complex System Simulation(LCSS)is often of large scale,and involves complex models and requires diverse resource.Consequently,it puts high demands on both the computational performance and usability of simulation platform.However,current cloud platforms mostly provide isolated simulation tools for users and consider neither requirements of easy assembly of simulation models nor the complex interaction between simulation entities,resulting in difficult usage and low execution efficiency.Therefore,it is of great theoretical and practical value to carry out researches on support technology of cloud simulation for LCSS in order to realize efficient integrated cloud simulation service to provide high efficient and easy-to-use service.Based on the current requirements of cloud simulation for LCSS as well as problems of related researches,this thesis focuses on the integration of serviced simulation tools,description of heterogeneous simulation resource,scheduling optimization of simulation resources,and fault tolerance of simulation execution.The main work and innovations are as follows:(1)An application-oriented integration framework for serviced simulation tools is proposed.Constructing an integrated cloud simulation service requires to realize both integration of serviced simulation tools and flexible assembly of simulation models.However,current cloud simulation platforms mostly provide isolated simulation tools,making it difficult to use.Besides,they require making modifications to support combination of existing simulation models.As a result,an integrated cloud simulation service cannot be provided.To solve this problem,an application-oriented integration framework for serviced simulation tools is proposed.The framework takes the application as the center and hides simulation tools in the integrated cloud simulation service,and uses an IDE simulation object structure which consists of initialization event,DR event,and functional event E to encapsulate simulation models in order to support flexible assembly of simulation models.Besides,XML is adopted to realize the transmission of information of simulation applications among serviced simulation tools.By this means,an integrated cloud simulation service is provided.Experiments show that this integration framework can provide support for realizing an easy-to-use integrated cloud simulation service.(2)A resource-retrieval-oriented semantic description method for heterogeneous simulation resources is proposed.The current resource description methods fail to implement semantic descriptions in combination with dynamic behaviors and historical interconnection records.As a result,the retrieval performance is not high.To solve this problem,a resource-retrieval-oriented semantic description method for heterogeneous simulation resources is proposed,and the ontology technology is used to semantically describe heterogeneous simulation resources from four aspects: basic information,dependency environment,usage context and dynamic information.Then,the logical matching and the similarity matching methods are used to realize separate and joint retrieval of simulation resources.Experiments show that OSRDM can effectively improve the retrieval performance of simulation resources.(3)A historical information based scheduling optimization algorithm of simulation resource is proposed.Combination schemes of virtual machines have great influence on simulation execution performance.Existing cloud resource scheduling methods are suitable for tasks whose sub-tasks are independent with each other and their running times can be determined in advance.However,simulation entities in a simulation application are frequently synchronized,and thus their running times are dependent upon each other and difficult to be determined beforehand.As a result,current resource scheduling methods are not suitable.To solve the problem,a historical information based scheduling optimization algorithm of simulation resource,i.e.PEMOA,is proposed.The method firstly uses the information of executed events generated during simulation execution to establish a simulation performance estimation model PEM,aiming to estimate the simulation execution time on different virtual machine combinations.Then,based on the results of PEM,the virtual machine combination scheme with the shortest running time is searched by adapting the genetic algorithm.Typical experiments show that PEMOA can reduce simulation execution time by up to 31.8%.(4)An online simulation fault tolerance method based on local collaboration is proposed.In cloud environment,the fault tolerance is often achieved based on replicas or checkpoints.However,the former needs to periodically maintain the state synchronization of replicas and brings about extra overhead;the latter needs to roll back fault-free simulation entities,and the simulation recovery efficiency is not high as a result.To solve this problem,an online simulation fault tolerance method based on local collaboration is proposed.This method adds simulation recovery execution module to the traditional simulation execution framework,and utilizes event buffering,message recognition and retransmission/filtering mechanism to realize,that only failed simulation entities are required to be restored to achieve simulation recovery.Then,efficient fault tolerance is supported in the cloud environment.The experimental results show that the proposed method can provide correct simulation results and can improve the efficiency of simulation fault tolerance and has a good scalability.Based on the above research results and the parallel discrete event simulation engine developed by our team,an integrated cloud simulation platform for LCSS,namely SIMCloud,is realized.The public opinion simulation experiment shows that SIMCloud is easy to use and can help improve simulation execution efficiency.