Research On Load Prediction And Container Scheduling Technology In Container Cloud Environment

With the rapid development of modern technology and electronic information technology and the arrival of the 5G era,it not only triggers the upgrade of cloud computing technology and the generation of cloud-native technology system,but also promotes the iteration of software architecture.At the same time,the cloud service platform based on container technology is also facing the situation that the container scale is getting larger and larger,the dependency relationship is getting more complex,and the resource structure is more diversified,which makes the artificial load balancing adjustment and resource optimization scheduling become history.Therefore,how to efficiently and accurately manage and schedule the unbalanced load and unreasonable resource usage in the container cloud environment has become one of the key points that need to be researched.With the continuous development and wide application of container technology,the research should focus on the modeling of historical load data and the prediction algorithm based on it,container-based migration and elastic scheduling technology,in order to solve the problem of inaccuracy and untimeliness of the current prediction algorithm and the ineffectiveness of container scheduling technology.Therefore,this paper designs and implements load prediction and container scheduling techniques for container clouds to seek targeted solutions,taking into full consideration the characteristics of container cloud environments such as diverse resource types,resource status changing at any time,service container status,container dependency,and load correlation.In this regard,the main research works of this paper are:(1)A load prediction method based on CNN-Bi GRU-Attention model is designed.Firstly,we analyze the requirements and problems faced by the load prediction technology in the container cloud field,and on this basis,we design a load model for container cloud resources,which combines static resource conditions and service dynamic factors to describe the load state;then we construct a CNN-Bi GRU-Attention model,which uses CNN and Bi GRU to capture the load time series data respectively.In order to improve the prediction effect,an attention mechanism is also introduced in the model to enhance the attention to the important information in the data to avoid the information loss due to the long sequence.Finally,the results of the load prediction method are analyzed to support the orderly execution of subsequent container scheduling.(2)In order to achieve a load-balanced and resource-optimized container cloud environment,a Container Scheduling Strategy Based On Load Prediction Model(CSSLPM,Container Scheduling Strategy)is designed.The strategy first uses the load prediction model to predict the load state,and then performs targeted coarse and fine grained container scheduling on this basis to ensure the optimal state of the container cloud environment as much as possible.On the one hand,it obtains the required number of container replicas in the next time window based on the scaling demand and performs fine-grained elastic scheduling based on the expected replica calculation method based on dynamic weights.On the other hand,on the basis of obtaining the migration demand and service container state,the coarse-grained migration scheduling based on service container state,container dependency and load correlation is realized by determining the containers to be migrated and the target nodes to be migrated through the established domain-based container selection method and Pearson correlation coefficient-based target node selection method;subsequently,the pre-mergebased container online migration The online migration process of the set of containers to be migrated is realized by using a pre-merge-based container online migration algorithm.(3)Based on the public dataset cluster-trace-v2018,the open source cloud computing simulation platform software Cloud Sim and the train booking benchmark system Train Ticket developed by Prof.Peng Xin’s team at Fudan University and open source,we designed corresponding experiments to test and improve the load prediction method and container scheduling strategy designed in this paper,and proved the accuracy and effectiveness of the proposed method.