Section | ||||
---|---|---|---|---|
|
Section | |||
---|---|---|---|
|
Author: The Fedora Development Team
...
A good discussion of the Fedora digital object model (for Fedora 2 and prior versions) exists in a recent paper (draft) published in the International Journal of Digital Libraries. While some details of this paper have been made obsolete (e.g. Disseminators) by the Content Model Architecture, a refinement of the original Fedora concepts introduced in Version 3.0, the core principles of the model remain the same. The Fedora digital object model is defined in XML schema language (see The Fedora Object XML - FOXML). For more information, also see the Introduction to FOXML in the Fedora System Documentation. A data object in a Fedora repository describes content (data and metadata) and a set of associated behaviors or services that can be applied to that content. Data objects comprise the bulk of a repository.
Figure 1 below shows the basic data model of a Fedora digital object.
Figure 1: Fedora Digital Object Data Model
...
Datastreams represent the digital content that is the essence of the digital object (e.g., digital images, encoded texts, audio recordings). All forms of metadata, except system metadata, are also treated as content, and are therefore represented as Datastreams in a digital object. All Datastreams have the potential to be disseminated from a digital object. A Datastream can reference any type of content, and that content can be stored either locally or remotely to the repository system.
A Datastream is the element of a Fedora digital object that represents a content item. A Fedora digital object can have one or more Datastreams. Each Datastream records useful attributes about the content it represents such as the MIME-type (for Web compatibility) and, optionally, the URI identifying the content's format (from a format registry). The content represented by a Datastream is treated as an opaque bit stream; it is up to the user to determine how to interpret the content (i.e. data or metadata). The content can either be stored internally in the Fedora repository, or stored remotely (in which case Fedora holds a pointer to the content in the form of a URL). The Fedora digital object model also supports versioning of Datastream content (see the Fedora Versioning Guide for more information).
...
Decisions about what to include in a Fedora digital object and how to configure its Datastreams are choices as you develop content for your repository. The examples in this tutorial demonstrate some common models that you may find useful as you develop your application. Different patterns of Datastream designed around particular "genre" of digital object (e.g., article, book, dataset, museum image, learning object) are known as "content models" in Fedora.
Figure 2: Fedora Digital Object Datastreams
...
...
Below is an alternative view of a Fedora digital object that shows the object from an access perspective. The digital object contains Datastreams and a set of object properties (simplified for depiction) as described above. A set of access points are defined for the object using the methods described below. Each access point is capable of disseminating a "representation" of the digital object. A representation may be considered a defined expression of part or all of the essential characteristics of the content. In many cases, direct dissemination of a bit stream is the only required access method; in most repository products this is only supported access method. However, Fedora also supports disseminating virtual representations based on the choices of content modelers and presenters using a full range of information and processing resources. The diagram shows all the access points defined for our example object.
For the access perspective, it would be best if the internal structure of digital object is ignored and treated as being encapsulated by its access points. Each access point is identified by a URI that conforms to the Fedora "info" URI scheme. These URIs can be easily converted to the URL syntax for the Fedora REST-based access service (API-A-LITE). It should be noted that Fedora provides a several protocol-based APIs to access digital objects. These protocols can be used both to access the representation and to obtain associated metadata at the same access point.
Figure 3: Fedora Digital Object Access Perspective
...
Fedora digital objects can be related to other Fedora objects in many ways. For example there may be a Fedora object that represents a collection and other objects that are members of that collection. Also, it may be the case that one object is considered a part of another object, a derivation of another object, a description of another object, or even equivalent to another object. For example, consider a network of digital objects pertaining to Thomas Jefferson, in which scholarly works are stored as digital objects, which are related to other digital objects representing primary source materials in libraries or museums. The composite scholarly objects can be considered a graph of related digital objects. Other types of objects can also be related to the scholarly object over time, for instance annotations about the scholarly object can be created by others and related to the original object. Also, digital objects can be created to act as "surrogates" or "proxies" for dynamically produced web content such as an Amazon page for a book relevant to the scholarly object. Such a network of digital objects can be created using Fedora, which in the abstract, would look like Figure 4.
Figure 4: A Network of Digital Objects
Digital object relationship metadata is a way of asserting these various kinds of relationships for Fedora objects. A default set of common relationships is defined in the Fedora relationship ontology (actually, a simple RDF schema) which defines a set of common generic relationships useful in creating digital object networks. These relationships can be refined or extended. Also, communities can define their own ontologies to encode relationships among Fedora digital objects. Relationships are asserted from the perspective of one object to another object as in the following general pattern:
...
<MyCatVideo> <is a member of the collection> <GreatCatVideos>
The creation of Fedora digital object relationship metadata is the basis for enabling advanced access and management functionality driven from metadata that is managed within the repository. Examples of the uses of relationship metadata include:
Object-to-Object relationships are stored as metadata in digital objects within a special Datastream. This Datastream is known by the reserved Datastream identifier of "RELS-EXT" (which stands for "Relationships-External"). Each digital object can have one RELS-EXT Datastream which is used exclusively for asserting digital object relationships.
A RELS-EXT Datastream can be provided as part of a Fedora ingest file. Alternatively, it can be added to an existing digital object via component operations of the Fedora management service interface (i.e., addDatastream). Refer to the FOXML reference example to see an example of the RELS-EXT Datastream in context. Modifications to the RELS-EXT Datastream are made via the Fedora management interface (i.e., modifyDatastream). The RELS-EXT Datastream is encoded as an Inline XML Datastream, meaning that the relationships metadata is expressed directly as XML within the digital object XML file (as opposed the relationship metadata existing in a separate XML file that the digital object points to by reference).
Fedora object-to-object metadata is encoded in XML using the Resource Description Framework (RDF). The relationship metadata must follow a prescribed RDF/XML authoring style where the subject is encoded using <rdf:Description>, the relationship is a property of the subject, and the target object is bound to the relationship property using the rdf:resource attribute. The subject and target of a relationship assertion must be URIs that identify Fedora digital objects. These URIs are based on Fedora object PIDs and conform to the syntax described for the fedora "info" URI scheme. The syntax for asserting relationships in RDF is as follows:
...
Yes! The Fedora repository service automatically indexes the RELS-EXT Datastreams for all objects as part of the RDF-based Resource Index.
This provides a unified "graph" of all the objects in the repository and their relationships to each other. The Resource Index graph can be queried using RDQL or ITQL which are SQL-like query languages for RDF. The Fedora repository service exposes a web service interface to search the Resource Index. Please refer to the Resource Index documentation for details.
...
A major goal of the Fedora architecture has been to provide a simple, flexible and evolvable approach to deliver the "essential characteristics" for enduring digital content. Whenever we work with digital content, it is with an established set of expectations for how an intellectual work may be expressed. With experience we develop "patterns of expression" that are the best compromise we can craft between the capabilities of our digital tools and the intellectual works we create in digital form. We store our digital content with the expectation that all the important characteristics of our intellectual works will be intact each and every time we return to access them, whether it has been a few minutes or many years.
...
By combining these very different views, CMA has the potential to provide a way to build an interoperable repository for integrated information access in our organizations and to provide durable access to our intellectual works. As we introduce CMA concepts, we will discuss the rationale behind the design decisions. This is only the first generation of the CMA and, like the rest of Fedora, we expect it to evolve. An understanding of design decisions behind this "first-generation" CMA is a key element for community participation in future generations of CMA development. Most important is an understanding of three significant and interrelated developments in software engineering: (1) object-oriented programming, (2) design patterns, and (3) model-driven architectures. It is beyond the scope of this document to discuss any of these developments in detail but we will make reference in this document to aspects of them which inform the design of the CMA.
The Content Model Architecture (CMA) describes an integrated structure for persisting and delivering the essential characteristics of digital objects in Fedora. In this section we will describe the key elements of the architecture, how they relate, and the manner in which they function. The original motivation for the CMA was to provide a looser binding for Disseminators, an element of the Fedora architecture used to stream a representation to a client. However, the CMA as described in this document has encompassed a far greater role in the Fedora architecture, in many ways forming the over-arching conceptual framework for future development of the Fedora Repository.
...
We must have a unique, unambiguous method to identify the class. For this purpose, in the CMA we have defined the "Content Model object," a digital object that both represents the notion of the class and can contain the content model. We use the identifier of the content model object (or CModel) as the class identifier. Following the rules of Fedora identifiers, the identifier of the CModel object can be encoded within a URI. We will describe the rationale for this decision in a later section but this approach provides two immediate benefits: (1) it provides a scheme which works within the Fedora architecture with minimal impact, and (2) it is compatible with the Web architecture, RDF and OWL. We can even build functionality using just the knowledge of the identifier without creating a content model. Having a uniform method for identifying a digital object's class maximizes interoperability.
The CMA does not require that digital objects explicitly conform to its architecture or explicitly declare any of its metadata elements beyond providing well-formed Fedora digital objects - unless you want to use the advanced features provided by the Fedora repository. The CMA uses a "descriptive" approach where the Fedora Repository will issue a run-time error for any operation it cannot perform. In most cases, you should still be able to disseminate bitstreams exactly as they are stored. CMA's dissemination approach is more consistent with the Web architecture, and provides a better balance between durable access to content and future innovations.
The minimum requirement to participate in the CMA is for a digital object to assert a relation to record its class' identity. Digital objects that do not explicitly identify their class are assumed to belong to a system-defined "Basic Content Model" which has a repository-defined reserved identifier. In CMA, you should use a digital object (see CModel below) as a way to register a "class" in a repository federation.
The remaining functionality is enabled by creating the content model and storing it within Fedora digital objects like any other content. This permits applications or the repository itself to disseminate the content model, interpret it and use it to provide functionality for the set of objects it describes. Content designers are free to develop content models (or even content modeling languages). Content Models and Content Model objects are designed to be shared. Digital objects having the same content model automatically form an interoperable information community. If the whole Content Model object is shared, keeping the object's identifier the same across multiple repositories, the community can easily extend across multiple organizations.
There are two basic approaches to using content models. First, the content and content models can be disseminated to applications able to interpret them to deliver the essential characteristics of the content. Disseminating the content model to external services can be also be used when creating new digital objects for ingest, validation, transformation and replication.
Alternately, the Fedora Repository can be used as a content mediation layer which interprets the content model to disseminate the content correctly. To accomplish this, the Fedora Repository must have access to code compatible with the Content Model. The compatible code may be added to the Fedora Repository using its plug-in architecture in combination with service mechanisms. While all the tools needed to plug in your own Content Model mediation are not complete in Fedora 3.0, this feature of the CMA will permit serving several generations of digital objects in the same repository reducing the need to update objects whenever there are new releases of Fedora. This will increase the durability of collections and enable longer periods between migrations of Fedora digital objects to new formats.
While the CMA does not force you to use a specific content modeling language, Fedora 3.0 contains a reference implementation that enables the Fedora Repository to operate
much as it did in prior versions. The following sections describe CMA in more detail and provide instructions on how to use the reference content modeling language so you can create your own CMA compatible objects immediately. Over time Fedora Commons will support the development of one or more content modeling languages as part of solution bundles that may be used by the community with minimum effort.
One of the basic elements of the Fedora architecture is the Fedora digital object. Every digital object stored in a Fedora repository is some variation of the same basic Fedora digital object model. In Fedora, digital objects containing data (Data object) utilize an XML element called the Datastream to describe the raw content (a bitstream or external content). In Fedora 2 and prior versions, digital objects containing data may also have contained Disseminators. The Disseminator is a metadata construct used by Fedora to describe how a client can access content within the digital object or remotely referenced by the digital object. If you only needed to access the raw content, default functionality was provided by the Fedora Repository which did not require that a Disseminator be explicitly added to the digital object. Unfortunately, the older design meant that the Disseminator was repeated in every Data object.
In Fedora 2 and prior versions, three specializations of the Fedora digital object were used: the Data object, the Behavior Definition (BDef) and the Behavior Mechanism (BMech). The BDef contained a description of the access interface which, in turn, was implemented by the BMech. In combination, the Fedora digital object model provides a uniquely flexible way to persist and access digital content. However, these features were not as simple to use as they could be. In particular, repeating the Disseminator in every Data object made changes to the access interface very difficult since every object had to be changed.
In the CMA, the Fedora digital object remains the model for all digital objects just as it was in prior versions of Fedora. However, for the CMA we define four specialized variations of the Fedora digital object (see Table 1).
Object Type | Code | Description |
---|---|---|
Data | Data | A container for content |
Service Definition | SDef | A container for service definitions, an element of a content model |
Service Deployment | SDep | A container for the service deployment bindings |
Content Model | CModel | A container for content models |
Table 1: Fundamental Fedora Object Types
Each of these specialized digital objects will be described in much greater detail below. The BDef and BMech have long been recognized as having reserved functions in Fedora
and we often called them "control" objects because they contain data used to control object-specific functionality. CMA adds a new control object, the CModel. However, some of the data needed for object-specific functionality was located in the Data object (often just labeled generically as the digital object) in prior Fedora versions. In particular, Disseminators were defined in the Data object. The CMA eliminates the Disseminator and redistributes control data in a more logical and reusable fashion among the "control" objects.
Care should be taken in equating the names of object types in the Fedora 3.0 from the prior (Release 2 and earlier) Fedora architecture because there are some similarities to roles played by these objects in both the old and new architecture. However, the CMA is a major redesign so care should be taken not to automatically equate the roles of the control objects in the old and new implementations. In pre-release reviews with community members, we discovered that there were concerns about the potential of confusion so we have used a new naming scheme for Fedora 3.0.
Also, care should be taken not to equate the digital object as a container and any data (or metadata) it contains. Since the deployment, functional, persistence and information views of the Fedora architecture tend to intermingle more than in most architectures, it is easy to accidently conflate characteristics of the architectural elements across views. We will try to make the distinctions clear where needed.
Table 2 lists the supported relationships between fundamental object types in the CMA.
Origin | Target | Relation | Purpose |
---|---|---|---|
Data | CModel | hasModel | Identifies the class and, optionally, the object containing a model of the essential characteristics of the class |
CModel | SDef | hasService | Identifies the object containing a model of the functional characteristics of class members |
SDep | SDef | isDeploymentOf | Identifies the object containing a model of the functions being deployed |
SDep | CModel | isContractorOf | Identifies the object containing a model of the information being deployed |
Table 2: Relationships Between the Fedora Object Types
Figure 5: Fundamental CMA Relationships
Figure 5 illustrates the required relationships between fundamental object types in the CMA. In the CMA object serialization these relations are asserted as RDF statements in the digital objects' RELS-EXT Datastream. These relations are asserted only in the object at the origin of each arrow though typically these relations will be harvested and indexed within utilities such as Semantic Triplestores or relational databases to enable fast query over them, or into caches which permit rapid access to their functionality.
The "hasModel" relation identifies the class of the Data object. There may or may not be a Fedora digital object that corresponds to the identifier. If the identifier refers to an object it must be a CModel object and contain the base content model document. It is expected that many Data objects conform to a single Content Model (and have a relation asserted to the same CModel object). The Content Model characterizes the Data objects that conform to it.
The SDef object describes a Service and the Operations it performs. Defining a Service is the means by which content developers provide customized functionality for their Data objects. A Service consists of one or more Operations, each of which is an endpoint that may be called to execute the Operation. This approach is similar to techniques found in both object-oriented programming and in Web services. The CModel object uses the "hasService" relation to assert that its' class members provides a Service (and its associated Operations). A CModel is free to assert relations to more than one Service. A Service may be related to many CModels.
Deployment of a Service in a repository is accomplished by using the "isDeploymentOf" relation to the SDef object. The Service Deployment (SDep) object is local to a Fedora repository and represents how a Service is implemented by the repository. Finally, the SDep object asserts the "isContractorOf" to indicate the CModel (effectively the class of Data objects) for which it deploys Services. This permits the SDep to access the Datastreams in the Data object and user parameters when an Operation is called for a Data object.
...
Thus far, we have talked about the component parts of a Fedora repository, but the larger picture is also important. A repository is made up of digital objects, but in what context do those objects exist and how is it that users interact with them?
Figure 6: Fedora System Architecture (simplified)
This diagram shows in very general terms the structure of the entire repository. Users interact with the content of the repository by means of client applications, web browsers, batch programs, or server applications. These applications access the repository's data by
means of the four APIs by which Fedora is exposed: management, access, search, which are exposed via HTTP or SOAP, and the OAI provider API, which is exposed via HTTP.
This diagram gives another view of the larger context of a Fedora repository. Users perform common tasks such as ingesting objects, searching the repository, or accessing objects via client applications or a web browser. These client applications mediate this interaction with the repository via web services on the frontend, and on the backend, the repository interacts with web services to perform any data transformations that are requested by users. The transformed data is then passed back to the user via the frontend web services.
It is important to note that users only interact with the repository via the APIs, even though it may sometimes seem that they are interacting directly with an object, they are not.
Figure 7: Client and Web Services Interaction