Research On Double Auction Based Intelligent Resource Allocation And User Bidding Algorithm For Cloud Environment

In recent years,with the wide application of the Internet and big data technology,big data analysis and processing has provided a supporting foundation for scientific research and the development of the industry.Cloud computing technology has developed rapidly due to its advantages of less dependence on user-side software and hardware.And more and more research institutions and enterprises have migrated service applications or computing resources to cloud platforms.Cloud computing enables organizations to reducing theirs operational and maintenance costs of purchasing and deployment of computer infrastructure.At present,cloud providers mainly use a fixed price mechanism to charge based on the size of the computing resources(cloud virtual machine)they provide,that is,ondemand instances.And users control the use of cloud computing resources according to their actual needs.At the same time,cloud computing resource providers will also sell some idle resources at dynamic price(Spot Instances),so the dynamic pricing mechanism based on bidding is also widely used in resource allocation algorithms in cloud environments.In this paper,we proposed a double auction mechanism,on the basis of single-side auction bidding,to let cloud providers also participate in the bidding auction process.And we will explore algorithms for more efficient and sustainable pricing and resource allocation in the cloud ecosystem.For the scenarios of single-type virtual machine configuration and heterogeneous virtual machine configuration,we model separately and use different matching and pricing algorithms to complete resource allocation,and then compare and analyze the performance including resource allocation efficiency of the proposed algorithm and traditional algorithm in the simulation environment.On the other hand,for the bidding optimization problem when users use spot instances,this paper starts with the price prediction of bidding instances,and proposes an imporved temporal convolutional network model that can better utilize historical data to predict future prices,and optimize the prediction algorithm to improve the prediction accuracy,so as to provide a basis for users to accurately bid when using Spot Instances.