Design of Microservice Application Performance Analysis Framework for Container Cloud
The automatic tuning technology of artificial intelligence model can provide high performance intelligent services of cloud data center at lower resource cost.However,artificial intelligence models and hardware devices are heterogeneous,and automatic tuning operations in cloud data centers will generate a lot of computing time,occupy computing resources,and generate energy consumption costs.To solve this problem,this paper designs an artificial intelligence model automatic optimization framework for cloud computing data center.An artificial intelligence model candidate configuration item filtering method is proposed,which uses the techniques of model construction,feature extraction,candidate item exploration and configuration query to resample the candidate search space and replace the inefficient candidate item with the efficient candidate item.At the operator optimization level,the framework performs hardware measurements implemented by compute components in batches in parallel,avoiding continuous probing of the search space.At the level of model optimization,the optimization of computing components across clusters is prioritized according to the relative performance acceleration of multiple AI models.The framework is designed to reduce the reasoning delay of AI models for different AI models and reduce the energy consumption of cloud computing data centers,thereby improving the cost-effectiveness of automatic tuning of AI models.
cloud data centerartificial intelligencemodel optimizationconfiguration exploration
万方数据
维普
