Design And Implementation Of Continuous Deployment Solution For Open Source CFD Cloud Platform

Computational Fluid Dynamics(CFD)is an interdisciplinary subject that uses computer technology and numerical algorithms to perform numerical simulation analysis of fluid forces.The spatial area occupied by the moving fluid is called the flow field.CFD software can provide researchers with the functions of flow field analysis,calculation and prediction,and its core module is the solver.PHengLEI is my country’s first industrialgrade open source CFD software created under the National Numerical Wind tunnel(NNW)project.NNW aims to support the needs of independent technological innovation in the development of key domestic aircraft and overcome the ”stuck neck” problem of CFD software.At present,PHengLEI is facing problems in multi-version solver parameter management,interoperability between platforms and solver deployment in the development process.To this end,we conducted research on:(1)To address the problem of how to efficiently match the solver with user requirements,this paper proposes an adaptive algorithm for CFD solver parameters.By comparing different versions of parameter files,the parameter change relationship is obtained and written in the file.By comparing different versions of the parameter file and writing them into the file,we can find the parameter change relationship according to the solver version number and the parameters in the case,and then complete the adaption.To further improve the lookup speed,we design a cache that combines hash tables and double-linked tables,and form a multi-level cache with Redis and Mysql.The feasibility of this algorithm is verified by successfully solving 6 cases in Windray Cloud platform using 10 solvers respectively; compared with the traditional Mysql management parameter approach,the lookup speed is improved by about 15 times on average.(2)Aiming at the problems that the automated test platform and the cloud platform are isolated from each other,resulting in inconvenient management and inefficient use of information resources and functional services,this paper designs and develops an integrated collaborative management platform based on the service-oriented architecture theory.The platform integrates the results of continuous integration in the automated test platform and the existing achievements of the cloud platform,and focuses on realizing the needs of users to customize the development of CFD software systems based on solvers.The main functions include: user management,data visualization,and CFD software system management.Through the actual application test and performance test of the platform,and the comparison and analysis of the test results,the efficiency and practicability of the platform are verified.(3)For the mainstream containers such as Docker due to its own privilege elevation and resource isolation mechanism and HPC scenarios there is a conflict problem.In this paper,we design an automatic deployment policy for CFD solver based on Singularity container technology,and develop a container management module,including container deployment and job submission functions.Finally,the automatic deployment strategy is functionally tested to verify its correctness.Then the performance test of the containerization scheme is conducted on Beijing supercomputers to verify the feasibility and effectiveness of the scheme.Through the research on the above three aspects,a set of solutions for CFD solver is formed,which provides strong support for my country’s independent CFD resear Sch and development.