Comparison between Chunk-Based and Layer-Based Container Image Storage Approaches: an Empirical Study

Container management frameworks, such as Docker, pack applications and their complex software environment in self-contained images, which facilitates application deployment, distribution, and sharing. Currently, container images are organized in a multi-layer file-system manner. For example, Docker uses AUFS to organize its container image storage. However, when a new version is created, all the changes are stored in a new layer via a Copy-On-Write mechanism, which can trigger a heavy burden for the network and storage. There are also other image storage approaches such as chunk-based technology. In this paper, we suggest that the container image storage system should not use one storage technology for all images, but should choose different technology for concrete images according to the image's distinct features. We design some experiments to compare the layer-based and chuck-based technology. And based on the analysis of the collected data, this paper gives some principles for the selection of image management techniques and discusses some interesting potential research topics.

I. Introduction

In the cloud computing era, the emergence of container technology has changed the process of software developing. Instead of configuring the environment from scratch, developers just need to pull the project image and start a container based on it. When they finish their developing, a new version of the image will be saved and published such that the service can be easily deployed [1]. Therefore, as the iteration of developing goes on, the image will be updated again and again. Currently, container images are organized in a multi-layer file-system manner. For instance, Docker [2] uses AUFS [3], a union file-system, to organize its image storage. When a container is created, a new layer of AUFS will be created and all the changes the user makes will be stored in this new layer.