The Research And Implementation Of Docker Image Distribution Technology For Data Center

Docker,a container-based virtualization technology,has been used widely in many corporations and datacenters.In large-scale datacenters,there can be hundreds of thousands of Docker containers are booted daily.It is important to increase the booting speed of Docker containers.If datacenters cannot boot quickly Docker containers which run applications to serve users,the quality of experience of users will be suffered.The storage and the distribution of Docker images play important roles when booting Docker containers,and have attracted much attention from both academia and large industry vendors.Several methods which use P2 P techniques have been proposed to accelerate the booting speed of Docker containers.However,these methods adopt simply P2 P file-downloading techniques without fully taking according the characteristic of Docker images and Docker registry.In this work,based on the characteristics of Docker image,we propose an efficient Hybrid Docker Image Distribution system for Datacenters(HDID).Further more,we develop a novel P2 P file-downloading strategy,which optimizes BitTorrent for Docker image distribution in datacenter networks.In particular,the main contributions of this paper are as follows:(1)The design and implementation of a hybrid Docker image distribution system for datacentersThe default Docker image distribution method use a node as registry server which manages and distributes images to clients.This method is not scalable and the registry node may become the single point of failure.To tackle this problem,the existing P2 Pbased Docker image distribution method simply treat a Docker image as a file to download,which does not take into account the characteristics of Docker images.In this work,we first make statistics and analysis the features of Docker image and Docker image layer.Then,based on the statistical results,we propose an efficient Hybrid Docker Image Distribution System for Datacenters,which embeds a widely used P2 P file distribution protocol into the Docker image repository.HDID selects dynamically the client-server method or P2 P method to distribute each layer of Docker image according to the size of each layer.Through extensive experiments,we show that the proposed method can reduce the image distribution time of Docker by about 90% compared to the default Docker image distribution method.(2)The design and implementation of a new Docker-based P2 P distribution strategy for datacentersTo further increase the booting speeding of Docker containers,we propose the Least number of Pieces First(LPF)strategy which improve adapt the traditional P2 P file-downloading strategies which was designed for the Internet instead of the datacenter.Specifically,LPF improve the peer-selection algorithm and the piece-selection algorithm of the original P2 P file-downloading strategies.For the peer-selection algorithm,based on the fact that there are no free-riders in datacenters,to reduce the maximal downloading time of all peers,LPF prefer hosts with fewer data pieces.For the piece-selection algorithm,unlike the Internet,nodes in datacenters stop rarely uploading pieces to others,we use random piece-selection algorithm instead of the rarest-first algorithm used by the original P2 P file-downloading strategies.Experimental results show that the LPF strategy proposed in this paper improves the speed of Docker images distribution in datacenters significantly compared with the traditional P2 P protocol algorithm and other state-of-the-art methods for datacenters.