Design And Implementation Of Decentralized Docker Image Resource Search Subsystem

With the development of computer technology,containers have become a fundamental building block of infrastructure of clusters.More and more operation and maintenance developers are also using container technology for development and deployment.The traditional image distribution method is to distribute images by a centralized image hub.This method cannot avoid the problems of performance bottleneck and node failure,and then P2P image distribution methods based on BitTorrent protocol appears,which can increase the bandwidth utilization of the whole network.However,these distribution methods based on BitTorrent protocol cannot completely avoid centralized resource index components,and the risk of a single point of failure still exists.In order to make the image distribution system work better in a more complex network environment,it is necessary to decentralize the search method of image resources and improve the distribution efficiency of the system as much as possible.For the above problems,this paper improves the distribution method,and designs a decentralized Docker image resource search subsystem.The research includes the following two parts:(1)Design a storage and indexing method of image resources based on the Kademlia network.The Kademlia network supports distributed key-value indexing.By analyzing the image data format,this paper uses the Kademlia network as the foundation to design the storage and indexing methods of image resources,and the interaction protocol to support the above indexing method.(2)Design a caching strategy for distributed indexes in the subsystem.In order to improve the efficiency while searching image resources,this paper adds a cache module to the distributed index module of the node.After analyzing the access mode of the image layer data,it is found that the cache efficiency can be improved by distinguishing the access mode of different resources at different nodes.After extending the protocol of the Kademlia network,a node can perceive the distance between the resource and the node,and through this distance parameter,the weights of the global probabilistic access mode and the strong local access mode for the resource can be quantified.After that,based on LRFU,which can integrate time-sensitive and frequency-sensitive access modes,this paper designs a caching strategy to optimize the efficiency of searching image resources.(3)Design and implement a decentralized Docker image resource search subsystem.Based on the search method of image resources designed in(1),design the overall architecture of the system,divide the system into module functions,design and implement the corresponding functional modules.Besides,based on the key technical design of the above theory,this paper built an experimental environment to test the performance of index caching strategy proposed in(2).The results proved that the proposed index caching strategy can improve the efficiency of decentralized image resource search.