Docker

Docker has emerged as the de facto standard application container engine [27]. Like virtual machines (VMs), application containers enable the encapsulation of software components so that any compliant computer system can execute them with no additional dependencies [18]. The advantage of Docker and similar container engines over virtualization consists of eliminating the need to run an operating system (OS) for each isolated environment. In contrast to hypervisors, container engines leverage kernel namespaces to isolate software environments, and thus run containers straight on the host OS. This makes application containers considerably lighter than VMs, enabling a more granular compartmentalization of software components.

By enabling the encapsulation of entire software stacks, container engines have the potential to considerably simplify application delivery. Engines such as LXC [74] and Jails [75] have been available for almost 2 decades. Nevertheless, when compared to Docker these systems are poor in terms of software delivery functionalities. This is the reason why software containers' popularity exploded only when Docker emerged.

Docker containers can be defined using a text specification language. Using such language, users compose a Dockerfile that is parsed by Docker and then compiled into a Docker image. Docker images can then be released to public or private registries, becoming immediately available over the Internet. Therefore, by running the Docker engine, the end users can conveniently start the released containers locally.

When using Docker containers for data processing, volumes play an important role. Indeed, there is a need for a mechanism to pass the input data to the containers and to retrieve the processed output from the isolated environment. Docker volumes allow for defining shared file spaces between containers and the host OS. Such volumes can be easily created when starting containers, by specifying a mapping between host OS file, or directories, and container mount points. Inside the containers these shared objects simply appear as regular files, or directories, under the specified mount point.