Scaling Identifiers and their Metadata to Gigascale: An Architecture to Tackle the Challenges of Volume and Variety

Persistent identifiers are applied to an ever-increasing variety of research objects, including software, samples, models, people, instruments, grants


INTRODUCTION AND MOTIVATION
The number of research artefacts that are identified through Persistent Identifiers (PID) and catalogued is rising rapidly.PIDs support a globally unique and unambiguous digital identification of resources on the Web.They have been around for over two decades (Klump & Huber 2017) and are now recognised as essential components of global research data ecosystems (Cousijn et al. 2021).Initially designed for scholarly literature and related resources, their application has expanded to various components of the scholarly ecosystem, including datasets, physical samples, 1 software, instruments, organisations, people, grants, and projects.Increasingly, identifiers are being used to track and link the research artefacts, starting with creating a digital resource (e.g., a dataset) through various stages of the research process -analysis, generation of further data, images, publication, and ultimately, its curation and preservation.In addition, there is a growing tendency to apply identifiers at a finer granularity and to use identifiers to link assets, such as through parent-child relationships.For example, not only can a collection of analyses have an identifier, its datasets and individual analyses within a dataset can be uniquely identified.
Physical samples play a key role in many disciplines, and often they serve as anchors in the physical world for data and its interpretation.To make these artefacts findable and accessible and to make their related data interoperable and reusable requires that this information is made available on the internet.PIDs for samples serve as this anchor.For example, physical samples can be linked to the virtual world by their digital representations (e.g., Lannom et al. 2019) and can be referenced by globally unique Web-resolvable persistent identifiers (e.g., Hardisty et al. 2021;Klump et al. 2021).The growing number of PIDs for samples, their description in online catalogues, and the required interoperability between these systems introduce three challenges: very large numbers of samples present the challenge of volume, while the many different use cases and ways to describe samples introduce the challenges of variety and variability.These dimensions were recognised early on as key challenges in the processing of large data volumes (Laney 2001).Other authors defined additional Big Data dimensions such as veracity and velocity (see e.g., Suresh 2014).However, velocity and veracity are not primary challenges faced by persistent identifier systems and federated catalogues.Therefore, in this paper, we will only discuss how we address the challenges of variety, variability, and volume.
For this paper, we will use physical samples and the International Generic Sample Number (IGSN) as an example of how the scaling problem in identifiers and metadata can be addressed.The IGSN was originally developed as the International Geo Sample Number to address the need for unambiguous identification of samples in geochemical databases and synoptic studies (Klump et al. 2021;Lehnert et al. 2004) but has since expanded in scope and is now used across many other disciplines.This paper will not discuss the IGSN in detail but will refer to Klump et al. (2021), where the technical details of the IGSN identifier, its underlying technology, and its governance are discussed.
To deal with the potential expansion of scale, the IGSN 2040 project (Lehnert et al. 2021) conducted a workshop in Canberra (Australia) to explore how metadata and sample catalogues could be shared at the scale of billions of samples (Klump et al. 2020).The aim of this workshop was to design a future system architecture for the minting and management of IGSN PIDs that would be able to scale to large numbers of samples, adapt to ever-changing applications for sample descriptions, and offer more vectors for discovery and services based on distributed sample catalogues.The outcome of the design process and discussions was a Web architecture-based approach that can accommodate both very large numbers of samples as well as thematic breadth across disciplines and use cases.Even though the IGSN technical development progressed in a different direction (Buys & Lehnert 2021), our proposal for a system architecture still has a broad application and is transferable to other research data infrastructures and PID systems.
The adoption of a new technology by a community depends heavily on how easy the technology can be adopted.The demise of the Life Science Identifier (LSID) showed that a technology 1 Different communities have their specific definitions of the terms 'specimen' or 'sample' to describe the sometimes complex relationships between a physical object of study and the 'thing' it specifically represents (see e.g., Haller et al. 2019).In this paper, we will use the term 'sample' for readability and follow its more common use in the geosciences community.stack that is difficult to implement will ultimately fail to be adopted by the community (Klump & Huber 2017).

THE CHALLENGE OF VOLUME
In research areas like geology, biology, or archaeology, it is common that thousands of samples are analysed before the data are then compiled into a dataset and described in a publication, which in turn, can then be aggregated in synthesis studies.It is, therefore, a reasonable assumption that publications and data are underpinned by orders of magnitude larger numbers of samples (e.g., Hardisty et al. 2021;Lannom et al. 2019).There is growing evidence that the scale at which identifiers are being applied in the research ecosystem is rapidly moving from the millions to billions, that is, Gigascale.The growing adoption of PIDs for individual samples across disciplines made many aware of the need to accommodate very large numbers of identifiers.In their scoping report, the Distributed System of Scientific Collections (DiSSCo) consortium estimates that natural history collections hold approximately 1.5 billion objects (Hardisty et al. 2021).This growth in numbers would put considerable strain on current architectures for minting, managing, publishing, finding, and accessing identifiers, and most existing systems would not be able to scale.
The increase in volume could be accommodated by adopting scalable cloud-based infrastructures for minting and managing identifiers.However, tools commonly used for harvesting metadata associated with identifiers have not been able to scale, and experience with IGSN has already shown that most common implementations of the Open Archives Initiative Protocol for Metadata Harvesting (OAI-PMH, Van de Sompel et al. 2004) become prone to failures once the number of items harvested from catalogues goes into millions of items.

THE CHALLENGES OF VARIETY AND VARIABILITY
Adoption of PIDs across research communities introduces the challenge that the increasing diversity of use cases cannot be accommodated by unified metadata schemas that attempt to generalise across use cases but need to be applicable in diverse communities (e.g., Damerow et al. 2021;Davies et al. 2021;Genova et al. 2017).This diversity in metadata needs to be reconciled with the Findable, Accessible, Interoperable, Reusable (FAIR) Principles of Wilkinson et al. (2016).To be able to scale to large numbers, the metadata has to be made machineactionable (e.g., De Smedt et al. 2020;Devaraju & Huber 2021;Ganske et al. 2020).The expansion of scope, combined with the large increase in the number of samples, requires an approach to metadata that can accommodate many different applications from a wide range of disciplines.
Building consensus on metadata is hard work and expensive (Genova et al. 2017), and metadata, like any language, is a social process driven by context and purpose (Eco 1997;Parsons et al. 2022).In many systems, the definition of schemas for the description of samples is too narrowly focused on one use case.A single common schema cannot convey a meaningful description of a sample without running the risk of being either a large but sparsely populated collection of loosely defined elements or a small set of metadata unable to hold much information about the object it describes.This situation is a common challenge in Webbased applications and has therefore been addressed by search engine operators and others (Guha 2011;Noy & Brickley 2017).
In this paper, we propose an approach to metadata that is prescriptive with respect to protocols but at the same time open to new applications by prescribing only a high-level information model that uses well-known Web architecture patterns to connect between systems.This approach allows us to accommodate a greater heterogeneity and variety of metadata than in the prescriptive approaches common in many research data infrastructures today, which often end up being restrictive in the number of communities that can apply them.
We, furthermore, describe a governance framework that is designed to help user communities develop and codify their expectations towards metadata structures that allow them to build information aggregators and other value-added services to serve their use cases.
The paper is organised as follows: Section 2 specifies a set of design principles to guide our technological choices.In Section 3, using use cases for the application of IGSN as an exemplar, we define the actors in the IGSN system -so-called personas -to explore their interactions with the IGSN system in the context of the use cases.From this, in Section 4, we determine a Minimum Viable Product of the system architecture that allows us to guide and prioritise the development of its components.Section 5 discusses our design choices and the broader implications of our proposal.

GUIDING PRINCIPLES
Before we defined the architecture of the envisaged infrastructure for the IGSN 2040 project, we defined a set of design principles based on a high-level design manifesto to guide our technical choices.The aim was to build a system that was easily extensible, scalable, sustainable, and flexible to accommodate the requirements of new science communities for their samplebased research.
We defined the following statements in our design manifesto following the example of Ross et al. (2020): • Our domains deserve research-specific software.
• Diverse practices and limited resources require generalised software.
• Do one thing well with modular and federated software (but slice the pie thoughtfully).
• Open-source software supports open research and has other advantages (but is difficult to maintain).
• Invest in outreach and engagement.
To support diverse practices, the technical architecture has to be general in nature, which is best achieved by adopting common practices for sharing metadata on the web.
For many stakeholders, sample lifecycles are only one consideration amongst many.Thus IGSN needs to fit into existing technology stacks, and the development of bespoke new toolsets should be avoided.Additionally, the broad spectrum of actors in the samples community means the ability to scale is vital -it should be easy to get started by adopting the deep thinking into sample-based science that IGSN already offers while also making it possible to do something new that is relevant to a given community without holding them back in the future.

WEB ARCHITECTURE
Web architecture implementations are mature and optimised for large-scale interactions to support the scale of sample data expected.Additionally, the architecture can be equally implemented on commodity hardware or in dynamically scalable cloud infrastructure.
Enabling a semantic network on Web architecture brings key capabilities for addressing largescale sample identifier services.Web architecture approaches have demonstrated the ability to scale to billions of resources and requests.Additionally, this capacity is a commodity and one that can be dynamically scaled in response to changing needs.Given the limited control over federated resources on the internet, error handling is an integral aspect of the Web architecture to allow the graceful handling of errors and redirection and forwarding of these events.
To enable a more adaptive approach to metadata, the proposed architecture uses Schema.org (https://schema.org/)together with JavaScript Object Notation for Linked Data (JSON-LD, https://json-ld.org/).Schema.org is a structured markup that standardises HTML tags for creating rich search engine results.JSON-LD is a lightweight linked data format and is a representation of the RDF data model, which does not require the same high degree of syntactic and semantic standardisation that is required in XML-based technologies and does not require establishing a consensus across the entire system.It allows the inclusion of a wide range of already published expressive semantic vocabularies into the Web architecture approach described above.This approach is similar to how internet search works without establishing a semantic consensus across the entire internet and thus enables both scale and reliability of the system.
Early on, the major internet search engine operators realised that Web crawlers were not the most efficient way of indexing the entire internet and that they often missed indexing the content of dynamically generated pages.To make it easier for search engines to find relevant content, the major search engine operators introduced the sitemaps protocol (sitemaps.org2006).The Sitemaps protocol allows a Webmaster to inform search engines about URLs on a website that are available for crawling.While the sitemaps protocol has been successful for search engine operators, only a few projects or eResearch infrastructures have used this protocol.Examples of the use of sitemaps to aggregate research data are the Environmental Data Initiative (EDI) (Servilla et al. 2018) and Biositemaps (Dinov et al. 2008), although both services seem to operate no longer.
Despite the limited uptake of sitemaps.orgamong research data infrastructures, we propose that the sitemaps.orgprotocol can be used by information aggregators serving specific communities and their content providers.In this scenario, the PID registrar would be the curator of a list of domain-specific sitemaps that would allow information aggregators to build domain-specific search engines.

CONTEXT AND VOCABULARY
Whereas Web architecture brings scale and resilience, the semantic network brings context.This context provides an understanding of the meaning and network of connections and properties in human and machine-readable ways.Such context enables richer search and semantic relations to be provided.The context maps terms to Web-resolvable, and hopefully machine-readable, references.
As a foundation, these references may leverage Schema.org as a lightweight and broad upperlevel vocabulary to help address general discovery and upper-level concepts (e.g., Michel & The Bioschemas Community 2018) and contribute to making samples findable.Complementing this is the ease with which this approach can be extended with research discipline or applicationspecific vocabularies to provide richer context to the samples to aid in addressing the remaining FAIR Guiding Principles such as access, interoperability, and reuse.This extensibility is a design feature of Schema.org.
This result is a common set of minimal registration metadata that is ubiquitous across the range of focused community terms and concepts.This facilitates simpler publishing and innovation to address community needs by defining a required core that can be extended by various communities of practice.This results in a clearer onboarding process for agents, science communities, and facilities, connecting them with a broader set of use cases.Communities that have already invested in the development of semantic types can preserve and leverage that work through community-specific extensions of the Schema.orgcore (e.g., Thessen et al. 2019).
Linked data represents accessible and connected data and combining this with relationships among the data facilitates a linked data Web as envisioned by Tim Berners-Lee ( 2009).Further, structured data on the Web and data on the Web best practices provide connections with Schema.org and other Web architecture-based linked data and semantic Web patterns.This union of semantic networks and Web architecture is a key principle of the proposed architecture.

GOVERNANCE
The governance of a (sample) PID system covers three main components and their interactions to ensure the effective management and use of the PIDs: (1) the roles and the actions of the entities involved, (2) the practices they must adhere to, and (3) the technology underpinning the processes.
An architecture that leverages multiple entities built on Web architecture needs strong governance to enable coordinated and efficient operation among the various communities and maintain interoperability.Where standards and projects guide the technical implementation of a PID-based sample system architecture, its governance focuses on guiding principles, best practices around the principles and community requirements that are fed back to the system implementers.Specifically, things such as alignment to principles and functional purposes (e.g., requirements of the PID communities) and publishing guidance (e.g., identifier practice standards) are examples of value-adding activities provided by governance.
Guiding principles such as FAIR and Linked Open Data (LOD) are programmatic foundations, but the interfaces and functional needs of communities are equally important.These highly focused and practical needs represent the communities' goals.With guidance, these too can be expressed as principles, decoupled from the technical implementation.
Guiding the publishers of sample information to express and disseminate the sample records in a standardised way helps ensure their resources are connected to the community's desired goals.Another important use of this is to bridge the representation of data and the services and interfaces expressed by aggregators and employed by the sample community.
On a technical level, critical elements of a Web architecture approach such as the resolution of identifiers to resources, curation of sitemaps, core schema models, and patterns for community extensions to those core models are examples of crucial governance activities.Conversely, approaches to indexing or the development of interfaces to such indexes may vary based on community or sub-community needs and should be enabled by the architecture.
As part of the governance, the IGSN must also ensure its agents are accountable as information providers.Agents must ensure and maintain the quality and accuracy of the continued access to and adherence to standard procedures for PID metadata.

OUTREACH AND ENGAGEMENT
Sample communities are diverse, and therefore community engagement is vital to foster the adoption of new architecture and its services.Mediating and bridging between broad crosscutting principles and focused community principles is an essential value-adding service of governance.Consensus-based decision-making and coordination are required to get the existing sample communities and potentially interested parties to transition into the new architecture.Like other federated information infrastructures, IGSN is a socio-technical system, and its success relies on the implementers of its components.A system that is bespoke and difficult to implement will struggle to find adoption.Our hypothesis was that using standard Web architecture components would make the implementation of the system easier for the stakeholders involved.To evaluate this hypothesis, the IGSN 2040 project organised a code sprint that was conducted in June of 2020 led by the authors of this paper together with members of the IGSN community, who between them allocated persistent sample identifiers (IGSN IDs), maintain a registry of these identifiers and aggregated information into federated catalogues (Fils et al. 2020).The code sprint was designed to build a rapid prototype to test the proposed Web architecture approach and structured data on the Web patterns using synthetic data, test the scalability of the system and uncover barriers to the adoption of the proposed technology stack.
The sprint also allowed the exploration of potential use cases and interfaces for harvested data.Outcomes of the sprint were a draft publication guidance documentation, a description metadata schema, and examples to support expressing sample metadata (Fils et al. 2020).Additional documentation of the evaluated architecture can be found in GitHub repositories associated with this sprint (Fils 2021;Robertson et al. 2020;Schindler & Devaraju 2020).

USER CASES AND PERSONAS: WHO IS IN THE GAME AND WHY
The design of the proposed system architecture is based on the analysis of services conducted during the 2019 IGSN 2040 workshop in Canberra, Australia (Klump et al. 2020).After defining the design principles for the future IGSN system architecture, workshop attendees identified the personas involved in the IGSN ecosystem.Next, they developed three workflows for use of IGSN.Finally, the group identified the services needed to support these workflows and mapped the personas to them.The personas and their interactions with the system architecture are summarised in Figure 1.
In this section, we discuss the concept of personas and describe the various personas involved in the proposed IGSN system.We provide two exemplary use case workflows: (1) linking samples to publications and (2) obtaining samples for reuse.We will describe the actors involved in these two use cases through the concept of personas.

PERSONAS
To model the users and their interaction with the system, we used the concept of 'personas'.This concept is used in user experience design to represent the goals and behaviour of a group of users, rather than an individual, and describes the role a person or organisation may take (Lidwell et al. 2010).Personas help guide the exploration of the context of behaviours and relationships between actors.The primary personas involved in the two use cases presented in this paper are IGSN Registrar, IGSN Agent, Information Aggregator, Information Provider, and Information User.
Any entity can interact with the system in multiple roles and can thus be expressed through more than one persona.We will consider the personas individually based on their interactions with the PID system.

a. IGSN Registrar
The IGSN Registrar represents two roles, one technical and the other in governance.The IGSN e.V., in partnership with DataCite (Buys & Lehnert 2021), fills this technical role by (a) operating the service for minting and resolving IGSN PIDs and (b) providing the central clearing point for linking IGSN Agents with other personas who want to use the data and samples.In the proposed system, the IGSN Registrar would maintain the sitemaps used by the Information Aggregators.The governance role includes mediating decisions for determining protocols and schemas for communication between IGSN Agents and Information Aggregators.The IGSN e.V. carries out the social aspect of coordinating the community in adhering to standards and best practices for sample publications and meeting functional requirements.

b. IGSN Agent
The IGSN Agent (e.g., institutions, agencies) provides IGSN registration services to Information Providers for samples and sample collections.To register an IGSN, Information Providers submit metadata to an IGSN Agent, which includes the required IGSN metadata kernel and communitydeveloped descriptive metadata specific to the IGSN Agent.The IGSN Agents develop and maintain controlled vocabularies and schemas used to validate the provided metadata.Once the submission and validation processes are complete, the IGSN Agent submits the registration metadata to the IGSN Registrar to mint the IGSN.The IGSN Agents also maintain the landing pages (Web pages to which the IGSN resolves), which contain the sample metadata in human and machine-readable formats and register their sitemaps with the IGSN Registrar.In the proposed architecture, these pages will be embedded with JSON and will be used alongside sitemaps to support harvesting by Information Aggregators (this is currently supported by OAI-PMH).IGSN Agents also provide administrative services to the Information Provider persona, which may include authentication and authorization for metadata management, processing namespace requests, and transfer of custodianship of registered metadata.

c. Information Aggregator
The Information Aggregator is both a consumer and secondary service provider.They firstly harvest sample metadata from one or more IGSN Agents to compile the structured metadata provided by them into an index (i.e., catalogue).Secondly, these indices are then leveraged to provide value-added services such as spatial indexes, clustering, or other.Many of these valueadded services may be implemented to address the goals or functional needs of a particular community of practice.Information Aggregators interact with individual IGSN Agents to access their schemas and vocabularies.They connect with the IGSN Registrar to access sitemaps or validate an IGSN.Information Aggregators may be a third party, or IGSN Agents may also act in the role of an Information Aggregator.

d. Information Provider
Information Providers are the originators, producers or collectors of samples who document the information (metadata) about the sample, which then supports the ability of others to reuse the sample.Information Providers rely on IGSN Agents for cost-effective, persistent management of this metadata.Motivations include tracking samples through their lifecycle (from field to lab, analysis to publication, etc.), meeting the requirements of a funding agency or publisher, and supporting discoverability and access.Information Providers register sample metadata following the guidelines of the appropriate IGSN Agent for their community.Information Providers maintain the metadata over time, for example updating information to the IGSN Agent about the sample or adding DOIs for derived datasets.Information providers may also contribute to discussions on vocabularies and metadata schema requirements within their IGSN Agent community.Information Providers may also act as Information Users.

e. Information User
Information Users are looking for samples or information about samples.They may visit an Information Aggregator to search and discover samples based on criteria such as a specific location or sample type.Information Users may obtain an IGSN off a sample container or from a publication (generated by an Information Provider) and resolve the IGSN to evaluate the metadata in order to reuse the sample or verify information about it.Resolving the IGSN connects the Information User to the IGSN Registrar, leading to the landing page maintained by the IGSN Agent.The Information User may utilise resources provided by the IGSN Agent to understand the vocabularies and schema used on the landing page.Information Users may also become Information Providers by generating additional information which updates or augments existing sample information.

USER STORIES AND WORKFLOWS
Having defined the personas, we now explain how they interact in two example workflows: 1) The 'sampling to publication' workflow The workflow 'sampling to publication' (Figure 2) begins with a researcher (Information Provider) collecting samples to address his research questions.The samples are described, and the metadata are registered with an IGSN Agent.The IGSN Agent is responsible for registering the samples with IGSN identifiers through the registration service supported by the IGSN Registrar, which includes providing to the IGSN Registrar the minimum metadata required to register a sample and obtain an IGSN and making the complete, descriptive metadata records submitted by the Information Provider about the samples available for harvesting to Information Aggregators.The researcher drafts a manuscript that contains the interpretations of data and observations based on their samples.The researcher includes the IGSN identifiers of the samples in his publication.The identifiers are resolvable to the metadata landing pages maintained by the IGSN Agent.The reviewers of the manuscript (Information Users) verify the IGSN identifiers for the samples (manually or through an automated process).As part of the scientific review, they may also obtain richer descriptions of the samples from the metadata landing pages.Once the review is complete, the manuscript is published.
2) The 'sample for reuse' workflow In the workflow for obtaining a 'sample for reuse' (Figure 3), an entity, acting as an Information Aggregator, develops and hosts a sample catalogue by harvesting metadata from IGSN Agent(s).A researcher (Information User) then searches the catalogue or a data portal to discover samples that may be relevant to their work.They evaluate the metadata profiles created by the Information Providers to determine if the samples may fit their research needs.
Once they identify the sample(s) of interest, they use the information provided in the record to contact the custodian of the sample and request the sample or a subsample for further studies.

RESULTS: IGSN SERVICES AND MINIMUM VIABLE PRODUCT
We used the personas and interactions described in the workflows above to define the essential elements of the IGSN ecosystem (Figure 4) that are needed to mint identifiers; enable the discovery of samples; disseminate metadata that aid the interpretation and potential reuse of the samples; and govern the IGSN system.Furthermore, we derived the core services necessary for IGSN to be able to serve its purpose, that is, the minimum viable product (MVP).The MVP is summarised in Figure 4 with the solid boxes representing the components of the MVP while the outlined boxes represent elements that can be added at a later stage.The system can be split into three groups of services: • PID minting and resolution; • discovery, reuse and governance; and • administration.
PID minting and resolution continue to be based on the Handle System, either directly, through DOI or another Handle-based identifier system.Metadata syndication and aggregation use the metadata embedded in the landing pages based on JSON-LD and Schema.org.Harvesters are directed to the relevant Web resources through sitemaps, where the IGSN Registrar acts as a registry of sitemaps.These sitemaps can be tailored to be community or use case specific.The IGSN Registrar also hosts registries of the schemas and vocabularies used in the Schema.organd JSON-LD encoded metadata, respectively.
In the federated architecture outlined in the introduction, the role of the IGSN registrar becomes more that of a coordinator than of a regulator.In this scenario, IGSN does not maintain a central catalogue of all registered samples, but only mints and administers the identifiers.Instead of maintaining a central catalogue, the IGSN registrar acts as a reliable pointer to resources, such as a curated list of sitemaps, that can then be harvested by Information Aggregators.

DISCUSSION
IGSN is not the only system facing the challenge of accessing and indexing large and diverse metadata catalogues.Similar challenges are faced by trans-disciplinary systems like the DiSSCo (Hardisty et al. 2021), DataCite (Neumann & Brase 2014), Bioschemas (Michel & The Bioschemas Community 2018), and the European Open Science Cloud (EOSC) (Schwardmann et al. 2021).
The challenges can be mapped to the dimensions scale, heterogeneity, and adoption.
With the proliferation of physical samples collected for various studies worldwide, it became apparent that discovering them over the Web was going to be challenging.One way of making them findable is through the syndication of metadata and the compilation of catalogues.Syndication of sample catalogues for indexing in federated portals had been a feature of the original IGSN system architecture (Klump et al. 2021).In its original design, IGSN used OAI-PMH because the protocol allows the syndication of metadata catalogues in more than one metadata schema (Van de Sompel et al. 2004).At the time when it was conceived, OAI-PMH was an elegant, lean, and fault-tolerant protocol for exchanging catalogues of several thousand objects.For the sharing of metadata and catalogues, XML is fairly verbose, and its mark-up adds significant bulk to the data payload, which is manageable with catalogues containing thousands to tens of thousands of entries but becomes a significant burden once catalogues scale to millions or billions of entries.Experience in the operation of the IGSN system has already shown that standard implementations of OAI-PMH become prone to failures once the number of harvested records goes into millions of items (Klump et al. 2021).
While the self-describing nature of XML seems to be an elegant solution, it turns out to be less so in practical application.The enforcement of schema constraints makes XML-based applications rigid in their information model and prone to runtime errors caused by syntactical inaccuracies (Barkstrom 2010).The simplifications introduced by JSON-LD and Schema.orgmake implementations easier without losing semantic richness.
Harvesting metadata is not unique to research applications but is also a core function of search engines.Indexing the internet at large led to the development of lightweight encodings based on JSON-LD.In early 2017, Google unveiled Google Dataset Search (Noy & Brickley 2017).Google Dataset Search builds on Schema.org,an initiative by Google, Bing, and Yahoo! to create and support a common vocabulary for structured data markup on Web pages (Guha 2011).
In 2018 the US National Science Foundation (NSF) EarthCube Science Support Office (ESSO) at UCAR, through pilot work funded by the office, developed a test leveraging Schema.orgtype Dataset and structured data on the Web to address issues of FAIR data for geoscience (Lingerfelt et al. 2018).From this evolved the ESIP Schema.orgCluster, which develops and publishes guidance for leveraging this approach for datasets (Jones et al. 2021).Additionally, EarthCube is extending this work as GeoCODES (https://geocodes.earthcube.org/) to develop and provide a discovery portal and associated knowledge graph for the NSF geoscience data facilities.
A similar approach was taken by the Bioschemas community (Michel & The Bioschemas Community 2018).They noted that Schema.org is well suited for describing data repositories but lacks terms to describe other research entities, such as biological taxa, proteins, genes, and reagents, that are of interest to biosciences.An additional challenge was that metadata records were distributed across several repositories.This information is required for aggregators to harvest, parse, and index very large volumes of metadata (Thessen et al. 2018).
Scaling metadata to large numbers may appear as the primary challenge, but when we look at the use cases described in the section above, most use cases will not need to index the entire catalogue of samples on the internet.A scalable architecture must allow information aggregators to select only a subset specific to their application.Decentralising the architecture of cataloguing samples on the internet makes it easier to structure the system components into modules that can be combined and adapted as needed in a specific application.The MVP outlined earlier describes the minimum components and functionality needed for building applications based on distributed sample catalogues.
The other challenge is the thematic breadth of samples on the internet and the many ways to describe them in diverse and evolving use cases.The use of structured data on the Web provides a means to leverage web architecture approaches in the samples community.This approach has many benefits: it leverages existing developer tools, skills, and experiences and can be implemented on existing server architecture.Furthermore, it builds on existing vocabulary efforts (e.g., Albertoni et al. 2021;Guha 2011;Haller et al. 2019) and enables connecting samples in a semantic manner to a broader range of resources like institutions, people, and publications.
Landing pages can serve rich and diverse descriptions of samples.To make this content machineactionable, it is important that communities of practice agree on the content that they expect to be displayed on landing pages.This approach has been pioneered in the community of earth system model data repositories (Ganske et al. 2020).
The above examples show how leveraging web architecture patterns around structured data for the Web gives access to the semantic Web and ways to encode the context around data.This makes building a network across science disciplines far easier as they do not require universal consensus among all stakeholders.In addition, the use of web architecture allows third parties like Google, Bing, DataOne, and others to access and use metadata to provide offerings based on open, well-known architectures.The IGSN 2040 code sprint in 2020 showed that using common Web technologies that are also used in search engines makes this approach flexible and easy to adopt (Fils et al. 2020).
The system architecture outlined in this paper has a broader impact than providing a technical solution to the challenge of indexing billions of objects across heterogeneous and distributed resources.By leveraging existing technology and approaches, a larger community is enabled to engage and make more samples discoverable and usable.This addresses the challenge of accessing very large numbers of records.The simplified architecture makes it easier to develop tools and interfaces, allowing the presentation of samples and their information in a manner aligned with a given community's needs.Furthermore, a simplified architecture aids sustainability from both a technical and financial perspective.
The open architecture of making metadata available as structured data on the Web democratises the publishing of knowledge by enabling more communities to represent their knowledge expressed in a way that is fit for their purpose and not limited by the least common denominator of an overarching standard.The nature of structured data on the web provides the ability to apply semantic context to samples.This means richer discovery and information about resources, their uses in the past and potential future uses becoming more readily available.Semantically enriched descriptions contribute directly to the implementation of the FAIR Guiding Principles (Wilkinson et al. 2016) beyond the aspects of findability (F) and access (A).Providing metadata that is interpretable by machines also makes them more interoperable (I) and makes them available for reuse (R) (Plomp 2020).

Figure 1
Figure 1 Schematic overview of the proposed system architecture based on Web architecture design principles.Colours used for personas are IGSN Registrar (green), IGSN Agent (yellow), information provider/user (blue), and information aggregator (red).

Figure 2
Figure 2 Workflow of linking samples to a publication.

Figure 3
Figure 3 Workflow of reusing samples.