Research And Application Of High Concurrency Performance Based On Aircraft Scar Detection Service

Skin scars on the surface of an aircraft are a major cause of threats to flight safety.Therefore,it is an important task for airlines to detect skin scars during aircraft transfers and stops to ensure flight safety.At present,the more advanced detection method is the scar detection method based on computer vision.This method transmits the collected aircraft skin image to the scar detection service,and performs image processing based on artificial neural network to identify the type and location of the scar,thereby judging the damage.The degree of skin damage ensures flight safety.Due to the large number of maintenance aircraft and short maintenance time,this method requires a large amount of data transmission and calculation during image transmission and processing.Therefore,how to ensure the real-time and high-efficiency of the scar detection during the transfer and stop of the aircraft is very important.This article is aimed at a specific airline's actual project,during the transfer and stop,the real-time and high-efficiency requirements of the image recognition-based scar detection service,and the study of high-concurrency performance solutions.Since airlines have high data security and privacy requirements,they cannot use the cloud platform of the public network to publish high-concurrency services.Therefore,the paper starts from single-machine distributed computing,dynamic load balancing,service platform deployment,and multi-machine The work is carried out from a distributed expansion perspective,and the following tasks are specifically completed:1.Aiming at the deployment problem of supporting the efficient operation of the service platform,it is launched from two aspects:hardware deployment and software scheduling.A single-machine multi-gpu distributed processing scheme is designed,and a multi-gpu distributed processing scheme based on the multi-process model is proposed.In order to solve the discreteness of user requests,a dynamic adaptive encapsulation strategy based on business traffic is proposed to improve transaction processing efficiency.2.Aiming at the problem of low concurrency of detection services due to unbalanced load of cluster nodes in the task distribution process,a dynamic load balancing strategy based on cluster performance is proposed,and performance collection modules,algorithm modules,and load balancing modules are designed and implemented to achieve dynamic task distribution.To ensure load balance between nodes,thereby increasing the concurrency of scar detection services.3.Aiming at the expansion difficulties and the cumbersome problems of launching in the process of multi-machine distributed expansion,a multi-machine distributed expansion scheme based on containerization technology is designed to realize the horizontal expansion of detection services.And with Docker as the underlying containerization technology,a set of production construction framework for continuous delivery and automated management is designed.The paper uses concurrent performance testing tools to simulate the detection requests in the real environment,designs corresponding experiments for the first and second tasks,and conducts control tests.The effectiveness and superiority of the above solutions are verified by QPS(Queries Per Second)indicators,and After adopting the packaging strategy proposed in this article,the QPS of the scar detection service increased by up to 32.33%.