Research And Implementation Of Elastic Scaling Technology Based On Cloud Native Training Platform

Traditional computer training courses are usually limited by software,hardware,and location.Especially during the epidemic situation,they have become a very difficult point in teaching.As far as I am concerned,cloudification of training courses plays an extreme important role in application value and the elastic scaling as its core also counts.This article first analyzes the requirements of the cloud training platform and according to actual teaching needs divides the cloud training courses into three types:command line,web application and interface interactive type.at the same time,add the training self-test function to make sure that after completing the requirements analysis,the architecture design of the platform will start.Combined with the actual scenario,the system function is modularized to ensure the high cohesion and low coupling of each module.Then the cloud training platform will eventually be implemented based on cloud native applications.This article studies the existing elastic scaling schemes in order to achieve the high availability of the platform on the basis of cloud training platform.And Analysis shows that when cloud training is enabled for elastic scaling,the loading of new nodes and the pulling of image is too slow.In the following content this article will propose a set of solutions to these two problems.Aiming at the problem of slow loading of elastic scaling nodes,this article designs an elastic scaling based on the ARMA forecast time series.Above all uses the data drive of the training monitoring system to help the elastic scaling module complete the forecast,and then realizes the elastic scaling scheme mainly based on ARMA prediction with a threshold value,The elastic scaling module partially based on the elastic scaling scheme.realizes the pre-loading of the nodes and achieves the purpose of optimizing the loading time of the nodes.Aiming at the problem of slow pull of training images,a mirror preloading technology based on the LRU strategy is proposed.The LRU is used to record the most frequently used training images.Based on the elastic scaling model predicted by ARMA,the mirroring in the LRU cache is performed.Preloading further optimizes the startup speed of cloud training in elastic scaling.Finally,the elastic scaling module for cloud training is embedded into the cloud training system for integration to complete a set of highly available cloud training platform.In order to ensure the stability,availability and reliability of the cloud training platform,this article will conduct basic testing,functional testing and performance testing.In the test,1200 machines were used to concurrently request 1976232 times,and the success rate reached 92.68%.When 100 training sessions were started,compared with the non-flexible scaling scheme,the resource utilization increased by 44.44%.In the end,the cloud-native cloud training platform passed all the tests,showing that it is a high-performance,highly reliable and stable system.