Design Of Distributed Multi-agent Programming Model Based On Pi-calculus

The distributed environment is a research hotspot in the field of high-performance computing.It can utilize distributed machine resources to collaboratively complete large-scale computing tasks.However,the scale of a distributed system often changes due to the joining or exiting of the machine,which further causes the network topology structure to change.How to ensure that the distributed system quickly adapts to the new topology and continue to complete the tasks to be performed,which is a problem that cannot be ignored.Parallel model,as the main force of the development of distributed systems,provides an execution template for high-dimensional,high-volume tasks.With the transformation of distributed computing environment from isomorphism to heterogeneity,what kind of parallel model can be used to reach the full potential in distributed systems,and achieve higher execution efficiency with simple configuration and maintenance,which is still the main research direction at present.Pi calculus can reflect the change of communication topology structure according to the dynamic change of channel link.It is suitable for describing distributed system with variable structure and has powerful mathematical theory theorem.It is a powerful means to verify the correctness of parallel model.Mobile Agents can perform unimpeded migration between heterogeneous machines and autonomously complete a series of activities such as computing,retrieving and collecting information.Due to the intelligence of the mobile agent,the communication overhead generated by the data synchronization process is alleviated,and the efficiency of the distributed system to perform tasks is further improved.Based on the research of distributed system,mobile agent and Pi calculus,this paper proposes a distributed multi-agent model based on Pi calculus,namely PDMA model.Specific tasks are as follows:1.Determine the overall design of the model.It is proposed to be dual,dynamic and efficient as the core.Dual refers to triggering parallel processes between machines and machines.Dynamic refers to dynamically adjusting cluster scale according to task volume and dynamically adjusting task allocation ratio according to machine performance.Efficient refers to machine performance.The priority is assigned to the task reasonably.For the evaluation of machine performance,the double factor of stability factor and variation factor is adopted,which makes the evaluation value of performance more reasonable.2.Build the model and verify.The process form of the Pi calculus represents the logical components of the PDMA model,and relies on the communication between the components and the cooperation between the agents.The channels in the Pi calculus are dynamic,infinite,easy to change and persistent,making the process of data transfer more advantageous than the traditional way.Then,according to the weak mutual simulation theorem provided by Pi calculus,the PDMA model is verified to ensure that the model is correct at the theoretical level.3.Programming implements model.The NPict programming language is a parallel language defined by Pi calculus,which can implement the model of Pi calculus.Therefore,this paper uses the NPict language to implement the PDMA model.Firstly,the publishing and subscription functions provided by NPict are used to realize the process of registering tasks to the model.After that,the Agent is dynamically created according to the amount of tasks.Finally,according to the Agent operating rules and channel usage rules,the distributed multi-agent cooperation accomplishes the task.4.Clear experimental examples.PDMA model,BSP model and serial computing process are implemented by NPict language respectively.By comparing the task execution time of the PDMA model and the BSP model,the efficiency of the PDMA model at the execution level is illustrated.By comparing the execution span speedup ratio between the PDMA model and the serial model,the feasibility of the PDMA model at the parallel level is illustrated.Through comparison of experimental results,it can be found that the PDMA model is more efficient than the BSP model and the serial model in the face of large-volume data tasks.That is to say,the application of the Pi calculus and Agent technology to the distributed computing system can reduce the execution overhead of the task.