Dr
Dieter Kranzlmuller
(LMU Munich)
# Background:
The approach of infrastructure-as-code allows to
efficiently manage large infrastructures, for instance to support
FAIR data management. A canonical and machine-actionable
description of these infrastructures can itself be an item of
research and an essential component in handling reproducibility
challenges for the results achieved on the infrastructures. Such a
description would include all necessary steps and dependencies in
order to provision the infrastructure described in a machine-
actionable and human-readable way.
# Objective:
The textual representations of the infrastructures need
both to be handled in a FAIR spirit and to be compliant to common
DevOps quality management standards: On the one hand, they should
be Findable, Accessible, Interoperable and Reusable. This can be
achieved by annotating the infrastructure-as-code with rich
metadata, including a permanent identifier that can be used to
retrieve the textual representation. On the other hand, they
should be integrated into a workflow of continuous integration and
deployment and managed by a version control system.
#Method:
Guidelines and best practices to meet these requirements
are presented, derived from experience with several research
infrastructures hosted at the Leibniz Supercomputing Centre. The
infrastructures considered span from small, but specialized
systems to large-scale HPC clusters. Where beneficial, interviews
with administrators were conducted to enrich the guidelines with
their experiences.
#Results:
The major recommendation is to stick to those standards
and tools that are already existent and relevant for the
continuous and scalable management of infrastructures in an
industrial context. Ansible is an example for a well-established
configuration and provisioning tool that finds applicaiton in both
scientific and industrial service proliferation. DataCite as a
generic metadata scheme is well-equipped to meet the domain-
specific requirements with regard to infrastructure description.
Beyond that the crucial part is to integrate both domains with
qualified references and show their potential with a proof-of-
concept.
#Evaluation:
The guidelines are exemplified by a proof-of-concept
to provision a generic research data infrastructure which runs on
a kubernetes cluster. This infrastructure includes services to
harvest data providers, a search engine and several backend
services to facilitate common workflows such as bookmarking of
search results, data staging and platforms to process and analyze
the data. The infrastructure as well as the code deployed on it
are open source and can be used to reproduce the presented
findings.Conclusion: The adoption of the recommendations presented cannot
only make infrastructure setups citeable, but might boost best
practices and facilitate the federation of distributed services in
the context of science.
Summary
The approach of infrastructure-as-code allows to efficiently manage large infrastructures, for instance to support FAIR data management. The canonical description of these infrastructures can itself be an item of research and an essential component in handling reproducibility challenges with regard to the results achieved on the infrastructures. Such a canonical description typically includes a machin-actionable set of instructions to setup machines, services and seed data. The textual representations of the infrastructures need both to be handled in a FAIR spirit and to be compliant to common DevOps quality management standards. Guidelines and best practices to meet these requirements are presented, derived from experience with large-scale research infrastructures hosted at the Leibniz Supercomputing Centre. The major recommendation is to stick to those standards and tools that are already existent and relevant for the continuous and scalable management of infrastructures in an industrial context. The only modifications necessary of these can be found where scientific context adds requirements that typically cannot be found in a business context. The adoption of these recommendations cannot only make infrastructure setups citeable, but might boost best practices and facilitate the federation of distributed services in the context of science.
