| In recent years,microservice architecture software has been deployed on the Mobile Edge of Computing(MEC)side in a form closer to users.Because some services have high latency requirements and limited resources on the edge side.Therefore,it is very important to ensure the service quality of latency-sensitive microservices and make full use of limited edge computing resources.Based on the above problems,the research contents of this paper are as follows:To ensure the service quality of latency-sensitive microservices,it is necessary to predict the service load in advance and expand the microservice instance.Considering the characteristics of microservices,this paper proposes the M-ED-MUL load prediction model based on edge computing scenarios.The prediction process firstly verifies the stationarity of the load sequence,and adopts a linear prediction model for the stationary sequence.For non-stationary sequences,firstly,for the upstream microservice load sequence with microservice calling relationship,the correlation degree is calculated and sorted and filtered to reduce the dimension,and the items with the highest correlation degree are obtained.Then,the undecimated wavelet transform algorithm is used to decompose the load sequence to be predicted into approximate signals and detail signals that are more conducive to computational analysis.Finally,these sequences are jointly input into the M-ED-MUL prediction model.The MED-MUL prediction model utilizes the long short-term memory encodingdecoding network to predict the target load at multiple time steps,and uses the multi-head attention mechanism to mine the pre-and post-dependency of the load sequence.Finally,the load prediction method proposed in this paper and other methods are fully experimentally compared on Alibaba Cloud’s microservice data set to verify the effectiveness of the method in this paper.In order to efficiently utilize the limited edge side resources,this paper proposes a multi-edge cluster microservice optimization deployment system.The entire system is optimized to ensure the service quality of delay-sensitive microservices and utilize edge resources in a balanced and efficient manner,and manages different edge clusters in a unified manner.Inside the edge cluster,the Kubernetes platform based on preemptive,customized pre-selection,optimization and other service deployment strategies is used for unified management.Between different edge clusters,a centralized system is used for unified management.The centralized system can predict microservice load,migrate service instances across clusters,and efficiently utilize multi-edge cluster resources.Experiments show that the multi-edge cluster scheduling system can maximize the utilization of multiple edge cluster resources.Compared with the inherent deployment strategy of Kubernetes,the deployment strategy proposed in this paper effectively ensures the service quality of delay-sensitive microservices,reduces the SLA violation rate of microservices,and effectively reduces the service delay. |
