Research On Coupling Communication Modeling And Resource Allocation Algorithm Of Climate System Model CESM

Community Earth System Model(CESM)is a global climate model coupled with multiple physical fields,which can simulate the global meteorological and environmental changes on a large scale and effectively predict the climate.In the research field of climate,CESM is a very important climate research tool,which has wide commercial value.The main framework of CESM is CCSM(the community climate system model).On the basis of CCSM framework,many climate prediction software with coupling structures have been developed by scientific research institutions around the world.The framework includes many physical components such as atmosphere,ocean,land and sea ice,and a coupler CPL(Coupler)for data exchange.The CPL component and other physical components exchange boundary data of differential equations to realize the coupling calculation of boundary values.Each component of CESM model contains complex physical and chemical processes.Through dynamic calculation of climate data collected worldwide,global climate change can be simulated.CESM is a typical high-performance computing application because of its complex computing process and huge amount of data needed for computing.Because the climate system model CESM is highly dependent on computing resources,it is often deployed to high-performance clusters in scientific research activities.Within the model,each component has a complex coupling computing model,which brings great pressure on the communication,memory access and computing performance of the computing platform.These factors all affect the running performance of the whole earth system model CESM.At present,for the research focus of CESM,researchers focus on improving the physical dynamics model of a certain climate component in CESM or researching and developing a new component.Single component technology has made obvious progress,but multi-component coupling has become the performance bottleneck of the whole program.Because of the complex communication environment and concurrent communication functions in high-performance clusters,it is difficult to model the communication part,so there is little work to consider the communication overhead of CPL at present.Therefore,for the optimization problem of multi-component resource allocation,the existing researches are mostly oriented to the resource division of each sub-component,but few people pay attention to the coupler CPL which undertakes important computing and communication tasks.The main work of this paper includes:(1)Communication performance is an important parameter that affects the operation optimization of CESM in Earth system mode.This paper deeply analyzes the communication behavior between coupler and other components,and models the overall communication performance of coupler and other components through the mixed multiple regression analysis model proposed in this paper;(2)When the communication performance is ideal,the communication speed is affected by the network bandwidth,etc.,and the computing speed obstacle of the platform becomes the bottleneck of the computing performance of the whole program.Therefore,this paper proposes a collaborative scheduling method CPSAS with independent execution of couplers.In this paper,more resources are allocated to the coupler component which is responsible for the communication of climate system mode components,that is,all the allocatable resources are fixedly allocated to the coupler CPL,the remaining components participate in the scheduling allocation,and the communication time between the whole coupler and other components is accumulated into the overall execution time of the earth system model software.(3)In view of the poor network communication environment,at this time,the communication performance of the platform has become the bottleneck of the computing performance of the whole earth system mode.This paper proposes a collaborative scheduling method CPSAI in which couplers participate in the arrangement strategy with other components.When other components are executed in parallel with coupler CPL,the communication time between each component and coupler will be hidden in the calculated real time of each component,thus optimizing the whole program.The main contributions of this paper include:(1)In this paper,the communication performance between CPL component and other components in CESM is modeled,which can well describe the communication performance cost between each component and coupler CPL,and the lowest mean square error of fitting is 0.743;(2)Compared with the default model,the CPSAS algorithm proposed in this paper achieves 41.35% performance improvement in IB network environment,and 7.7% performance improvement compared with CPSA algorithm.The CPSAI algorithm proposed in this paper can improve the performance by 32.43% compared with the default model in Ethernet environment.(3)This paper compares and analyzes the collaborative scheduling method CPSAS of coupler independent execution and CPSAI of coupler participating arrangement,which are put forward in this paper,and their practical application in climate system mode.