CN-115935502-B - Pneumatic shape optimization design task scheduling method for supercomputer

CN115935502BCN 115935502 BCN115935502 BCN 115935502BCN-115935502-B

Abstract

A pneumatic appearance optimization design task scheduling method for a super computer belongs to the field of aircraft optimization design. In order to improve the working efficiency of an optimal design task of a supercomputer, a local workstation is used as a master control device of the optimal design task of the aerodynamic appearance, a high-performance computing cluster is used as an aerodynamic numerical simulation analysis device, the master control device of the optimal design task of the aerodynamic appearance initiates a parallel sample point set evaluation task, a first available CPU core of the master control device is used as a dispatching core, the dispatching core is responsible for decomposing a single sample point evaluation process in the parallel sample point set evaluation task, constructing a task queue system, controlling the sample point set evaluation process and generating a sample point set evaluation result file identifiable by the optimal design program, and other available CPU cores of the master control device are used as computing cores in the parallel sample point set evaluation task and are responsible for executing each sub-process of single sample point evaluation in the optimal design task. The invention greatly improves the sampling efficiency of the optimal design.

Inventors

YANG LONG
Wei chuang
SONG JIAYIN
ZHANG TIEJUN
QIAN ZHANSEN

Assignees

中国航空工业集团公司沈阳空气动力研究所

Dates

Publication Date: 20260505
Application Date: 20221109

Claims (5)

1. A pneumatic appearance optimization design task scheduling method for a supercomputer is characterized in that a local workstation is used as pneumatic appearance optimization design task master control equipment, a high-performance computing cluster is used as pneumatic numerical simulation analysis equipment, the pneumatic appearance optimization design task master control equipment initiates parallel sample point set evaluation tasks, a first available CPU core of the pneumatic appearance optimization design task master control equipment is used as a scheduling core, the scheduling core is responsible for decomposing a single sample point pneumatic evaluation process in the parallel sample point set evaluation tasks, constructing a task queue system, controlling the sample point set pneumatic evaluation process and generating a sample point set evaluation result file identifiable by an optimization design program, other available CPU cores of the pneumatic appearance optimization design task master control equipment are used as computing cores, and the computing cores are responsible for executing all sub-processes of the single sample point evaluation in the pneumatic appearance optimization design tasks in the parallel sample point set evaluation tasks, and comprise three sub-processes of sample point evaluation preprocessing, high-performance computing cluster invoking pneumatic numerical simulation analysis and sample point evaluation post-processing; The method comprises the following steps: S1, dividing all available CPU cores of the pneumatic appearance optimization design task master control equipment into a scheduling Core and a calculating Core, wherein the scheduling Core is the first available CPU Core, the calculating Core is other available CPU cores, the scheduling Core codes Core0, the calculating Core codes CoreX, X= (1, 2.., N-1), and N is the number of all available CPU cores; S2, the scheduling core reads in sample point set information to be evaluated of a parallel sample point set evaluation task, wherein the sample set information comprises the number of samples of the parallel sample point set evaluation task, the dimension of a single sample point, a sample point evaluation file and sample point evaluation script information; s3, setting the initial state of the computing core as idle, and creating a free_id_X.dat file, wherein no content exists in the free_id_X.dat file; s4, the scheduling core creates a single sample point pneumatic evaluation process execution folder according to serial numbers of single sample points in the parallel sample point set, copies a sample point evaluation file into the single sample point pneumatic evaluation process execution folder, then decomposes a single sample point pneumatic evaluation process script into three subprocess scripts of single sample point evaluation pretreatment, pneumatic numerical simulation analysis and single sample point evaluation post-treatment, and sets script serial numbers, wherein subprocess 1 represents the single sample point evaluation pretreatment process, subprocess 2 represents the pneumatic numerical simulation analysis process, and subprocess 3 represents the sample point evaluation post-treatment process; S5, the dispatching core searches whether an idle computing core exists, if so, a GoToWork _ID_X.dat file is written out, the free_ID_X.dat file is deleted, the GoToWork _ID_x.dat file content is a single sample point pneumatic evaluation process execution folder address created in the step S4, and if not, the dispatching core sleeps for 5 seconds; S6, searching whether the GoToWork _ID_X.dat file exists in the idle computing core, if so, obtaining a task by the idle computing core, and if not, continuing to wait until the GoToWork _ID_X.dat file appears or the optimization design is finished; S7, executing a sub-process corresponding to a sequence number in a GoToWork _ID_X.dat file in an address shown by the file by a computing core for obtaining a task, carrying out optimal design, executing sub-processes 1 and 2 by all the computing cores in the initial stage of the optimal design, executing a sub-process 3 by the computing cores in the later stage of the optimal design, determining the initial stage and the later stage of the optimal process according to the proportion of the number of sample points of the executed sub-processes 1 and 2 to the sample point set, generating a free_id_X.dat file by the computing cores after the task execution is finished, and returning to an idle computing core queue; s8, repeating the steps S4-S7 until all sample points are evaluated; and S9, the scheduling core generates a sample point set evaluation result summarizing file according to the requirements of the pneumatic appearance optimization design task, and the pneumatic appearance optimization design task for the supercomputer is completed.
2. The method for dispatching pneumatic appearance optimization design task of supercomputer according to claim 1, wherein the sample point evaluation file in the step S2 comprises a model generation script, a grid generation script and an initial grid file.
3. The method for dispatching the pneumatic appearance optimization design task for the supercomputer is characterized in that in the step S4, a single sample point pneumatic evaluation process script is decomposed into three subprocess scripts including single sample point evaluation preprocessing, pneumatic numerical simulation analysis and single sample point evaluation post-processing, wherein the method comprises a fixed sequence method, parameterization and a dynamic grid are set as sample point evaluation preprocessing subprocesses, a cluster task is submitted to be a pneumatic numerical simulation analysis subprocess, and result processing is conducted as a sample point evaluation post-processing subprocess.
4. The method for dispatching the pneumatic appearance optimization design task for the supercomputer according to claim 3, wherein in the step S4, the single sample point pneumatic evaluation process script is decomposed into three sub-process scripts including single sample point evaluation pretreatment, pneumatic numerical simulation analysis and single sample point evaluation post-treatment, wherein the given keyword method is adopted, the sample point pneumatic evaluation process is written by adopting a computer script language, and the sub-process decomposition is carried out by adding keywords into the single sample point pneumatic evaluation process script.
5. The method for dispatching the pneumatic profile optimization design task for the supercomputer according to claim 3 or 4, wherein the later optimization design stage is started when the number of sample points of the sub-process 1 and the sub-process 2, which are executed at the initial stage of the optimization process, in the step S7 is 95% of the sample point set.

Description

Pneumatic shape optimization design task scheduling method for supercomputer Technical Field The invention belongs to the field of aircraft optimization design, and particularly relates to a pneumatic appearance optimization design task scheduling method for a super computer. Background With the development of aviation technology, aircraft designs are evolving towards refinements and rapid iterations. The aerodynamic shape optimization design of the aircraft does not only consider aerodynamic characteristics, but also comprehensive design in multiple disciplinary directions such as aerodynamic, structural, stealth, flight mechanics and the like. The aerodynamic profile optimization design of the aircraft is characterized by multiple parameters, multiple constraints, multiple targets and multiple extrema, and higher requirements are placed on the efficiency and effect of the optimization design. In engineering, in order to achieve a satisfactory optimal design effect, a large grid quantity and high-precision algorithm are required to be adopted for carrying out numerical simulation evaluation on each subject of the aircraft, and a multi-round iterative optimization mode is adopted, so that the number of times of numerical simulation evaluation on each subject of the aircraft is large, the single evaluation time is long, and the optimal design efficiency is seriously affected. Therefore, a single workstation is often adopted to match with an supercomputer form to develop a fine design task. When a single workstation is matched with an supercomputing form to carry out an aircraft optimization design task, a synchronous parallel mode is adopted, namely a plurality of sample point evaluation processes in the optimization process are parallel, and a single sample point evaluation process is serial. The sample point evaluation process is a process from a set of design variables to obtaining optimization targets (responses), constraints corresponding to the set of design variables, such as in an aircraft aerodynamic profile optimization design, and includes parameterization, grid generation, numerical simulation analysis, and result extraction. The serial sample point evaluation process is to carry out parameterization, grid generation, super-arithmetic value simulation analysis operation submitting, (super-arithmetic) value simulation analysis and result extraction for a single core of a single workstation. When the supercomputer performs the numerical simulation analysis, the core submitting the supercomputer numerical simulation analysis operation on the single workstation is in a dormant state. When the single-machine workstation adopts a 4-core CPU to carry out the optimal design task, 4 sample point evaluation processes are carried out at most simultaneously, 4 calculation tasks are submitted at most on the super calculation, if the optimal design task needs to carry out 500 sample point evaluations (1 million grids and 192 super calculation cores execute pneumatic numerical analysis time of 1 hour), the parallel numerical simulation analysis time of the single-machine 4-core CPU is 5.2 days, which is unacceptable. Although the current domestic super-computing resource reaches the level of ten thousands of cores, the synchronous parallel mode cannot fully utilize the super-computing resource, is limited by the parallel cores of a single workstation, and reduces the working efficiency. Disclosure of Invention The invention aims to solve the problem of improving the working efficiency of an optimal design task of a super computer and provides a pneumatic appearance optimal design task scheduling method for the super computer. In order to achieve the above purpose, the present invention is realized by the following technical scheme: A pneumatic appearance optimization design task scheduling method for a supercomputer is characterized in that a local workstation is used as pneumatic appearance optimization design task master control equipment, a high-performance computing cluster is used as pneumatic numerical simulation analysis equipment, the pneumatic appearance optimization design task master control equipment initiates parallel sample point set evaluation tasks, a first available CPU core of the pneumatic appearance optimization design task master control equipment is used as a scheduling core, the scheduling core is responsible for decomposing a single sample point pneumatic evaluation process in the parallel sample point set evaluation tasks, constructing a task queue system, controlling the sample point set pneumatic evaluation process and generating sample point set evaluation result files identifiable by an optimization design program, other available CPU cores of the pneumatic appearance optimization design task master control equipment are used as computing cores, and the computing cores are responsible for executing all sub-processes of the single sample point evaluation in the pneumatic appearance o