Because of growing demand to preserve and share reproducible computational experiments in scienti c community, there has been interest in using Docker Linux Containers as a preservation mechanism. However, this is insu cient to help researches to comprehend "Dockerized" experiments and connect computational artifacts with concepts in peer-reviewed publications. We present here an ontology and software, Smart Container, that can conceptualize Docker artifacts by and is aligned with other existing vocabularies such as the well known W3C prov vocabulary.
Given the complexity of software systems, an initial ontology certainly cannot cover every single aspect related to Docker. Therefore, we focused on conceptualizing essential terms involving running a computational experiment in Docker importing existing ontologies and conceptual terms into an ontology pattern. The purpose of this work is to ll the provenance gap between Docker infrastructure and scienti c experiment artifacts and providing a vocabulary prototype that is capable for future extension. 2.1
The Smart Container Ontology was constructed from a systematic alignment between the main concepts present in existing Docker meta-data utilizing and existing vocabulary terms where possible to contextualize those meta-data concepts. In our work, PROV and CSO, two widely-used vocabularies, were introduced to construct the ground of Smart Container Ontology.
PROV-O is a W3C recommendation that describes the interactions of
provenance generated in di erent systems and under di erent contexts[
The Core Software Ontology(CSO)[
Docker is an application based on Linux Containers(LXC). It isolates an application with its dependencies in a single process which is more light-weighted than full hypervisor virtualization of guest operating system. It can be provisioned by a simple Docker le text based work ow. Docker also adopted a layer le system way to achieve versioning and component re-use. A Docker image is a read-only layer which is stateless. A container has states: when it is running, it represents a tree of processes isolated from other processes on the host; when it exits, it represents a read-write layer generated by the process along with its all underneath stateless images. We di erentiated these two concepts in our ontology. 4 http://purl.org/ifgi/dolce# 5 http://www.loa.istc.cnr.it/ontologies/ExtendedDnS.owl 6 http://www.loa.istc.cnr.it/ontologies/Plans.owl 7 http://www.ontologydesignpatterns.org/ont/dul/IOLite.owl 2.2
From inspection, a Docker image is a digital object with some attributes. It
matches the description of a prov:Entity: a physical, digital, conceptual, or other
kind of thing with some xed aspects[
In computational experiments, we have to be very careful about some special concepts from the computer science domain. The encoding of the whole computational experiment, such as the Docker le, is similar to InformationObject concept from CSO. The Docker itself in the experiment, on the other hand, is similar to a form of CSO:InformationRealization which is a realization of code in the machine. If we break the Docker le line by line, we also can treat each line of command as a smaller InformationObject. The running container is analogous to ComputationalActivity in CSO where the software manifests itself by a sequence of tasks contained in a plan. Our approach is Ontology Pattern based creating our own specialization but apply rdfs:seeAlso to terms in CSO, the weak sense of identity without making strong ontological commitments based on DOLCE.
In g 1, sc:Image, representing a Docker Image, is a specialization of prov:Entity. Docker containers were divided into three parts: sc:startContainer, sc:runningContainer and sc:endContainer. sc:startContainer and sc:endContainer subclass prov:Entity representing static conditions. sc:RunningContainer, on the other hand, subclasses prov:Activity as an event over time. An Docker le, referenced as sc:Docker le, and a line of command, referenced as sc:Command, both are subclasses of prov:Plan. The human user of the Docker le is identi ed as sc:User which subclassing prov:Person. sc:SoftwareAgent is a direct subclass of prov:SoftwareAgent standing for the software executes commands. We use a technique similar to TrustURI's by using the Docker image 64 digit code uniquely can be resolved by a uniform resource name(URN) with speci c protocols to create a URI for a static image. Each running container can be exposed by a HTTP address which is dereferenceable so we construct Uniform Resource Locator(URI) in the normal manner. We identify a human agent using URI's constructed from ORCID(Open Researcher and Contributor ID)identi er, a non-proprietary alphanumeric code to uniquely identify scienti c and other academic authors, facilitating investigator and potential publication identities to be propagated.
In this paper, we present an ontology pattern named Smart Container that contextualizes the docker software system acting as an infrastructure for computational experiments. We populated our ontology design pattern by analyzing main concepts in Docker and aligned with PROV-O and CSO to provide possibilities for wider extensions.
Acknowledgements. We acknowledge funding from NSF grant PHY-1247316 \DASPOS: Data and Software Preservation for Open Science."