The Object Data Management Standard: ODMG 3.0
Morgan Kaufmann, 2000, ISBN 1-55860-647-5
|ODMG 2.0 Book Extract|
|Note: This text is for ODMG 2.0. The new release, ODMG 3.0, was published in January 2000. The following text does not reflect the ODMG’s charter change. The ODMG is now an Object Storage API standard that can work with any DBMS or tool. We are in the process of changing all references to Object Databases (ODBMS) to Object Data.
|Table of Contents|
|2 Object ModelCHAPTERS
2 Object Model
2 Object Model
|2 Object Model
2 Object Model
4.3 Query Input and Result
4.4 Dealing with Object Identity
4.5 Path Expressions
4.6 Null Values
4.7 Method Invoking
4.9 Operator Composition
4.10 Language Definition
4.11 Syntactical Abbreviations
4.12 OQL BNF
5 C++ Binding
5.2 C++ ODL
5.3 C++ OML
5.4 C++ OQL
5.5 Schema Access
5.6 Example 6 Smalltalk Binding
A Comparison with OMG Object Model
This book defines the ODMG standard for object database management systems. It should be useful to engineers, managers, and students interested in object database systems. Although product documentation from ODMG member companies covers similar information, this book represents the definitive reference for writing code that will work with multiple products. The book has also proven useful in courses on object databases, and as an overview of object database programming.Release 2.0 differs from the previous Release 1.2 in a number of ways. With the wide acceptance of Java, we added a Java persistence standard in addition to the existing Smalltalk and C++ ones. We made the ODMG object model much more comprehensive, added a meta-object interface, defined an object interchange format, and worked to make the programming language bindings consistent with the common model. We made changes throughout the specification based on several years experience implementing the standard in object database products.With Release 2.0, the ODMG standard has reached a new level of maturity. Products now available comply reasonably well with the C++ binding, Smalltalk binding, OQL, and even the new Java binding. As with Release 1.2, we expect future work to be backward compatible with Release 2.0. Although we expect a few changes to come, for example to the Java binding, the standard should now be reasonable stable. For any changes since publication of this book, please check the contact information at the end of Chapter 1, particularly the web site www.odmg.org.
ODMG created a working group for each chapter of this book. The authors listed on the cover of this book are the elected chairs and editors for those working groups. Work on standards is not always recognized as important to companies whose livelihood depends on next quarter’s revenue, so these authors are to be commended on their personal dedication and cooperation in improving the usability and consistency of their technology. In addition, other people have made important contributions to the ODMG working groups and to previous releases of this standard. These people are acknowledged in Chapter 1.
Rick Cattell, March 1997
Chapter 1 [top]
We have worked outside of traditional standards bodies for our efforts in order to make quick progress. Standards groups are well suited to incremental changes to a proposal once a good starting point has been established, but it is difficult to perform substantial creative work in such organizations due to their lack of continuity, large membership, and infrequent meetings. It should be noted that relational database standards started with a database model and language implemented by the largest company involved (IBM); for our work, we have picked and combined the best features of implementations we had available to us.
1.1.1 Importance of a Standard
All of these factors are important for object DBMSs, as well. The scope of object DBMSs is more far-reaching than that of relational DBMSs, integrating the programming language and database system, and encompassing all of an application’s operations and data. A standard is critical to making such applications practical.
The intense ODMG effort has given the object database industry a “jump start” toward standards that would otherwise have taken many years. ODMG enables many vendors to support and endorse a common object database interface to which customers write their applications.
1.1.2 Goals [top]
We are striving to bring programming languages and database systems to a new level of integration, moving the industry forward as a whole through the practical impetus of real products that conform to a more comprehensive standard than is possible with relational systems. We have gone further than the least common denominator of the first relational standards, and we want to provide portability for the entire application, not just the small portion of the semantics encoded in embedded SQL statements.
The ODMG member companies, representing almost the entire ODBMS industry, are supporting this standard. Thus, our proposal has become a de facto standard for this industry. We have also used our specification in our work with standards groups such as the OMG and the ANSI X3H2 (SQL) committee.
We do not wish to produce identical ODBMS products. Our goal is source code port-ability; there is a lot of room for future innovation in a number of areas. There will be differences between products in performance, languages supported, functionality unique to particular market segments (e.g., version and configuration management), accompanying programming environments, application construction tools, small versus large scale, multithreading, networking, platform availability, depth of functionality, suites of predefined type libraries, GUI builders, design tools, and so on.
Wherever possible, we have used existing work as the basis for our proposals, from standards groups and from the literature. But, primarily, our work is derived by combining the strongest features of the ODBMS products currently available. These products offer demonstrated implementations of our standards components that have been tried in the field.
1.1.3 Definition [top]
In summary, we define an ODBMS to be a DBMS that integrates database capabilities with object-oriented programming language capabilities. An ODBMS makes database objects appear as programming language objects, in one or more existing programming languages. The ODBMS extends the language with transparently persistent data, concurrency control, data recovery, associative queries, and other database capabilities. For more extensive definition and discussion of ODBMSs, the reader is referred to textbooks in this area (e.g., Cattell, Object Data Management).
1.2 Architecture [top]
1.2.1 Major Components
It is possible to read and write the same database from C++, Smalltalk, and Java, as long as the programmer stays within the common subset of supported data types. More chapters may be added at a future date for other language bindings. Note that unlike SQL in relational systems, ODBMS data manipulation languages are tailored to specific application programming languages, in order to provide a single, integrated environment for programming and data manipulation. We don’t believe exclusively in a universal DML syntax. We go further than relational systems, as we support a unified object model for sharing data across programming languages, as well as a common query language.
1.2.2 Additional Components [top]
1.2.3 ODBMS Architecture Perspective [top]
The programmer writes declarations for the application schema (both data and interfaces) plus a source program for the application implementation. The source program is written in a programming language (PL) such as C++, using a class library that provides full database OML including transactions and object query. The schema declarations may be written in an extension of the programming language syntax, labeled PL ODL in the figure, or in a programming language-independent ODL. The latter could be used as a higher-level design language, or to allow schema definition independent of programming language.
The declarations and source program are then compiled and linked with the ODBMS to produce the running application. The application accesses a new or existing database, whose types must conform to the declarations. Databases may be shared with other applications on a network; the ODBMS provides a shared service for transaction and lock management, allowing data to be cached in the application.
1.3 Status [top]
It is to the personal credit of all participants that the ODMG standard has been produced and revised expeditiously. All of the contributors put substantial time and personal investment into the meetings and this document. They showed remarkable dedication to our goals; no one attempted to twist the process to his or her company’s advantage. The reviewers were also very helpful, always ready to contribute.
In addition to the regular ODMG participants above, we received valuable feedback from others in academia and industry. We would like to thank our academic reviewers, in particular, Eliot Moss for his contribution to the object model chapter. We would also like to thank Joshua Duhl for his exhaustive review of Release 1.2 as part of the process to create certification test suites.
1.3.2 History [top]
ODMG was conceived at the invitation of Rick Cattell in the summer of 1991, in an impromptu breakfast with ODBMS vendors frustrated at the lack of progress toward ODBMS standards. Our first meeting was at SunSoft in the fall of 1991.
The group adopted rules that have been instrumental to our quick progress. We wanted to remain small and focused in our work, yet be open to all parties who are interested in our work. The structure evolved over time. Presently, we have established work-groups, one for each chapter of the specification. Each workgroup is intended to remain small, allowing for good discussion. The specifications adopted in each work-group, however, must go before the ODMG Board for final approval. The Board usually holds open meetings for representatives from all members to attend and comment on our work.
The people who come to our meetings from our member companies are called Technical Representatives. They are required to have a technical background in our industry. We also have established rules requiring the same Technical Representatives come repeatedly to our meetings to maintain continuity of our work. Technical Representatives from voting member companies often contribute 25 percent of their time to the ODMG. Some of the Technical Representatives from our reviewer members contribute an equal amount of time while others are more in the range of ten percent of their time.
Voting membership is open to organizations that have developed and commercially sell a currently shipping ODBMS as defined on page 3. Reviewer members are individuals or organizations having a direct and material interest in the work of the ODMG. Certification members, our newest group of members, are individuals or organizations having a direct and material interest in the certification work of the ODMG.
1.3.3 Accomplishments [top]
1.3.4 Next Steps [top]
1.3.5 Related Standards [top]
X3 Secretariat, CBEMA
There are also references to Object Management Group (OMG) specifications, from the Object Request Broker (ORB) Task Force (also called CORBA), the Object Model Task Force (OMTF), and the Object Services Task Force (OSTF). OMG can be contacted at:
Chapter 2 [top]
Chapter 3 defines programming-language independent Object Specification Languages. One such specification language, Object Definition Language (ODL), is used to specify application object models and is presented for all of the constructs explained in this chapter for the Object Model. It is also used in this chapter to define the operations on the various objects of the Object Model. Chapters 5, 6 and 7, respectively, define the C++, Smalltalk and Java programming language bindings for ODL and for manipulating objects. Programming languages have some inherent semantic differences; these are reflected in the ODL bindings. Thus some of the constructs that appear here as part of the Object Model may be modified slightly by the binding to a particular programming language. Modifications are explained in chapters 5, 6, and 7.
The Object Model specifies the constructs that are supported by an ODBMS:
The ODMG Object Model specifies what is meant by objects, literals, types, operations, properties, attributes, relationships, and so forth. An application developer uses the constructs of the ODMG Object Model to construct the object model for the application. The application’s object model specifies particular types, such as Document, Author, Publisher, and Chapter, and the operations and properties of each of these types. The application’s object model is the database’s (logical) schema.
Analogous to the ODMG Object Model for object databases is the relational model for relational databases, as embodied in SQL. The relational model is the fundamental definition of a relational database management system’s functionality. The ODMG Object Model is the fundamental definition of an ODBMS’s functionality. It includes significantly richer semantics than does the relational model, by declaring relation-ships and operations explicitly.
Object Specification Languages
Two specification languages are discussed in this chapter: Object Definition Language (ODL) and Object Interchange Format (OIF).
3.2 Object Definition Language
Several principles have guided the development of the ODL, including:
ODL is not intended to be a full programming language. It is a definition language for object specifications. Database management systems (DBMSs) have traditionally provided facilities that support data definition (using a Data Definition Language – DDL) and data manipulation (using a Data Manipulation Language – DML). The DDL allows users to define their data types and interfaces. DML allows programs to create, delete, read, change, etc., instances of those data types. The ODL described in this chapter is a DDL for object types. It defines the characteristics of types, including their properties and operations. ODL defines only the signatures of operations and does not address definition of the methods that implement those operations. The ODMG standard does not provide an OML specification. Chapters 5, 6 and 7 define standard APIs to bind conformant ODBMSs to C++, Smalltalk and Java respectively.
ODL is intended to define object types that can be implemented in a variety of programming languages. Therefore, ODL is not tied to the syntax of a particular programming language. Users can use ODL to define schema semantics in a programming language independent way. A schema specified in ODL can be supported by any ODMG-compliant ODBMS and by mixed-language implementations. This portability is necessary for an application to be able to run with minimal modification on a variety of ODBMSs. Some applications may in fact need simultaneous support from multiple ODBMSs. Others may need to access objects created and stored using different programming languages. ODL provides a degree of insulation for applications against the variations in both programming languages and underlying ODBMS products.
The C++, Smalltalk and Java ODL bindings are designed to fit smoothly into the declarative syntax of their host programming language. Due to the differences inherent in the object models native to these programming languages, it is not always possible to achieve consistent semantics across the programming-language specific versions of ODL. Our goal has been to minimize these inconsistencies, and we have noted, in Chapters 5, 6 and 7, the restrictions applicable to each particular language binding.
The syntax of ODL extends IDL – the Interface Definition Language developed by the OMG as part of the Common Object Request Broker Architecture (CORBA). IDL was itself influenced by C++, giving ODL a C++ flavor. Appendix B, “ODBMS in the OMG ORB Environment,” describes the relationship between ODL and IDL. ODL adds to IDL the constructs required to specify the complete semantics of the ODMG Object Model.
ODL also provides a context for integrating schemas from multiple sources and applications. These source schemas may have been defined with any number of object models and data definition languages; ODL is a sort of lingua franca for integration. For example, various standards organizations like STEP/PDES (EXPRESS), ANSI X3H2 (SQL), ANSI X3H7 (Object Information Management), CFI (CAD Framework Initiative), and others have developed a variety of object models and, in some cases, data definition languages. Any of these models can be translated to an ODL specification (Figure 3-1). This common basis then allows the various models to be integrated with common semantics. An ODL specification can be realized concretely in an object programming language like C++, Smalltalk or Java.
Chapter 4 [top]
Object Query Language
We first describe the design principles of the language in Section 4.2, then we introduce in the next sections the main features of OQL. We explain the input and result of a query in Section 4.3. Section 4.4 deals with object identity. Section 4.5 presents the path expressions. In Section 4.7, we show how OQL can invoke operations and Section 4.8 describes how polymorphism is managed by OQL. Section 4.9 concludes this part of the presentation of the main concepts by exemplifying the property of operators composition.
Finally, a formal and complete definition of the language is given in Section 4.10. For each feature of the language, we give the syntax, its semantics, and an example. Alternate syntax for some features are described in Section 4.11, which completes OQL in order to accept any syntactical form of SQL. The chapter ends with the formal syntax, which is given in Section 4.12.
4.2 Principles [top]
Chapter 5 [top]
ODL stands for Object Definition Language. It is the declarative portion of C++ ODL/OML. The C++ binding of ODL is expressed as a library which provides classes and functions to implement the concepts defined in the ODMG object model. OML stands for Object Manipulation Language. It is the language used for retrieving objects from the database and modifying them. The C++ OML syntax and semantics are those of standard C++ in the context of the standard class library.
ODL/OML specifies only the logical characteristics of objects and the operations used to manipulate them. It does not discuss the physical storage of objects. It does not address the clustering or memory management issues associated with the stored physical representation of objects or access structures like indices used to accelerate object retrieval. In an ideal world these would be transparent to the programmer. In the real world they are not. An additional set of constructs called physical pragmas is defined to give the programmer some direct control over these issues, or at least to enable a programmer to provide “hints” to the storage management subsystem provided as part of the ODBMS runtime. Physical pragmas exist within the ODL and OML. They are added to object type definitions specified in ODL, expressed as OML operations, or shown as optional arguments to operations defined within OML. Because these pragmas are not in any sense a stand-alone language, but rather a set of constructs added to ODL/OML to address implementation issues, they are included within the relevant subsections of this chapter.
The chapter is organized as follows. Section 5.2 discusses the ODL. Section 5.3 discusses the OML. Section 5.4 discusses OQL – the distinguished subset of OML that supports associative retrieval. Associative retrieval is access based on the values of the properties of objects rather than on their IDs or names. Section 5.6 provides an example program.
5.1.1 Language Design Principles
Chapter 6 [top]
6.1.1 Language Design Principles
6.6 Future Directions [top]
A uniform set of Database administration operations would facilitate application port-ability and allow system administration tools to be constructed which could work uniformly across multiple vendors’ database products.
This binding has only touched upon the need for interface regeneration mechanisms. Such mechanisms would allow programmers with existing applications utilizing language-specific and even database-specific ODL mechanisms to produce the inter-face definitions which would insulate them from the differences between these mechanisms.
Chapter 7 [top]
7.1.1 Language Design Principles
Note that the Java binding provides persistence by reachability, like the ODMG Smalltalk binding (this has also been called transitive persistence). On database commit, all objects reachable from database root objects are stored in the database.
7.1.2 Language Binding [top]
One possible ODMG implementation that supports these capabilities would be a post-processor that takes as input the Java .class file (bytecodes) produced by the Java compiler and produces new modified bytecodes that support persistence. Another implementation would be a preprocessor that modifies Java source before it goes to the Java compiler. Another implementation would be a modified Java interpreter.
We want a binding that allows all of these possible implementations. Because Java does not have all hooks we might desire, and the Java binding must use standard Java syntax, it is necessary to distinguish special classes understood by the database system. These classes are called persistence-capable classes. They can have both persistent and transient instances. Only instances of these classes can be made persistent. The current version of the standard does not define how a Java class becomes a persistence-capable class.
Appendix A [top]
Comparison with OMG Object Model
The bottom line is that the ODMG Object Model (ODMG/OM) is a superset of the OMG Object Model (OMG/OM).
The subsections of this appendix discuss the purpose of the two models and how the ODMG/OM fits into the component/profile structure defined by the OMG/OM, and review the capability between the two models in the major areas defined by the OMG/ OM: types, instances, objects, and operations.
A.2 Purpose [top]
To offer real portability, a standard has to support a level of DBMS functionality rich enough to meet the needs of the applications expected to use the standard. It cannot define such a low-level API that real applications need to use functionality supplied only by vendor-specific extensions to the API. The low-level, least-common-denominator approach taken in the standards for relational data management has meant that real applications need to use functionality supplied only by vendor-specific extensions to the API. Several studies in the late 1980s that analyzed large bodies of applications written against the relational API (SQL) showed that 30-40% of the RDBMS calls in the application are actually “standard SQL”; the other 60-70% use vendor-specific ex-tensions. The result is that the relational standard does not in practice deliver the source-code-level application portability that it promised. The ODMG APIs have been designed to provide a much higher level of functionality and therefore a much higher degree of application portability.
A.3 Components and Profiles [top]
The ORB profile includes the Core Component plus support for remote operations.
The ODBMS profile includes the Core Component plus support for
It also strengthens the core component definition of operations by including exception returns.
To date, the only OMG/OM component that has been defined is the Core Component. The additional functionality included in the ORB profile has not been formally specified as a set of named components. Nor are there OMG component definitions for the functionality expected to be added by the ODBMS profile. One of the reasons for making the comparison between the OMG/OM (i.e., the Core Component) is that the members of ODMG expect to submit definitions of each of the items in the bulleted list above as candidate components, and the sum of them as a candidate profile for object database management systems. Since the submitting companies collectively rep-resent 80+% of the commercially available ODBMS products on the market, we assume that adoption of an ODBMS profile along the lines of that outlined in Chapter 2 will move through the OMG process relatively quickly.
The OMG/OM is outlined below, with indications how the ODMG/OM agrees.
Types, Instances, Interfaces, and Implementations:
The OMG/OM does not currently define exception returns on operations; it says that it expects there to be an exception-handling component defined outside of the core model. The ODMG/OM does define exception returns to operations.
A.4 Type Hierarchy [top]
The ODMG/OM is a richer model than the OMG/OM – particularly in its support for properties and in its more detailed development of a subtype hierarchy below the types Object and Literal. The only differences between the two models in the areas common to them are two type names. The type that is called Literal in the ODMG/OM is called Non-Object in the OMG/OM. Although the OMG/OM does not formally introduce a supertype of the types Object and Non-Object, in the body of the document it refers to instances of these two types as the set of all “denotable values” or “Dvals” in the model. In the ODMG/OM a common supertype for Object and Literal is defined. The in-stances of type Object are mutable; they are therefore given OIDs in the ODMG/OM; although the value of the object may change, its OID is invariant. The OID can therefore be used to denote the object. Literals, by contrast, are immutable. Since the in-stances of a literal type are distinguished from one another by their value, this value can be used directly to denote the instance. There is no need to ascribe separate OIDs to literals.
In summary, the ODMG/OM is a clean superset of the OMG/OM.
A.5 The ORB profile [top]
A.6 Other Standards Groups [top]
1. Application-specific standards groups that have defined an object model as a basis for their work in defining schemas of common types in their application domain, e.g.,
2. Formal standards bodies working on generic object models, e.g.,
It is our current working assumption that the OMG-promulgated interface definitions for ORB and ODBMS will have sufficiently broad support across software vendors and hardware manufacturers that interface definitions put in the public domain through the OMG and supported by commercial vendors will develop the kind of de facto market share that has historically been an important prerequisite to adoption by ANSI and ISO. Should that prove not to be the case, the ODMG will make direct proposals to ANSI and ISO once the member companies of ODMG and their customers have developed a base of experience with the proposed API through use of commercial ODBMS products that support this API.
Appendix B [top]
ODBMS in the OMG ORB Environment
B.1 Introduction [top]
This document is a position statement of the ODMG defining the desired architecture. It explicitly does not discuss the architecture of the internals of an ODBMS implementation but rather leaves that to the implementor of the ODBMS. Instead, it discusses how the ODBMS fits architecturally into the larger OMG environment.
The issues for a successful fit are the following:
The architecture must support ODBMS implementations and client interfaces to achieve these.
B.2 Roles for the ORB and ODBMS [top]
Note that we define ODBMS according to the services it provides, not according to any particular implementation of those services. Radically different implementations are possible, including not only traditional ODBMS approaches, but also file-based approaches, each offering different levels of services and trade-offs.
The ORB provides a larger-scale set of services across heterogeneous vendors and products; e.g., it allows clients to use multiple ODBMSs. The service it provides is behavior invocation, or method dispatch. In contrast, the ODBMS provides a single-vendor capability and only a specific set of services rather than arbitrary ones; however, those services include more detailed capabilities of high-performance, fine-grained persistence that are used directly within applications to support millions of primitive objects. The ORB, when it needs persistence services, could choose to implement them via use of an ODBMS. The ODBMS services may be invoked via the ORB.
B.3 Issues [top]
In many cases this means access time that is comparable to native in-memory object usage. To provide this, the ODBMS must be able to move objects as necessary in the distributed environment and cache them locally in the address space of the application, if desired, and in efficient format.
Since the ODBMS objects are those used primarily within the application, it is desirable to support an interface that is natural and direct to the user.
Examples of applications and object granularities for which ODBMS services must be available and efficient include spreadsheets; word processors; documents of these; primitive elements within these such as cells, formulae, variables, words, phrases, and formatting specifications; network managers with objects representing machines, users, and sessions; resource allocation schemes; CAD and CAM with objects such as circuits and gates and pins, routing traces, form features, bezier curves, finite element mesh nodes, edges and faces, tool paths, simulation, and analysis support; financial portfolio analysis; and so on. There may be millions of such objects, in complex inter-connected networks of relationships.
The interfaces to those objects must be defined in such a way as to allow ORB access when appropriate (e.g., for cross-database-vendor object relationships) or direct use of the ODBMS (e.g., for objects with no need to publish themselves for public use through the ORB). This should be done with a single interface to allow transparency to the client and to allow the client to choose to vary functionality as desired.
The ODBMS acts as manager of many objects, so the architecture and interfaces must allow such assignment of responsibilities. The ODBMS can provide distribution of objects among multiple and potentially heterogeneous platforms, so the architecture and implementation must allow this functionality to be relegated to the ODBMS.
The ORB and other OMG components (service providers, library facilities, service users, etc.) may need the services of persistence, or management of objects that exist beyond process lifetimes, for various kinds of objects, including type-defining objects and instances of these. It is desirable, architecturally, to consolidate common services in a common shared component. The architecture must allow use of an ODBMS for this purpose in order to take advantage both of the capabilities it provides and integration with other OMG components using the same services.
As mentioned above, different ODBMS implementation approaches must be supported. The architecture and the OMG interfaces must provide a single interface (or set of interfaces) that allows use of a wide variety of such implementations. A single inter-face allows users to choose which implementation to use and when. This should cover not only full ODBMS implementations but other approaches with partial functionality, such as file management approaches.
In addition to direct use of an ODBMS through an interface such as that defined in the ODMG-93 specification, an ODBMS could be decomposed in order to implement a number of semi-independent services, such as persistence, transactions, relationships, and versions. The OMG Object Services Task Force is defining such services. This is an area for future work by the object database vendors. In addition to the ORB and users of the ORB accessing ODBMSs, it is also the case that an ODBMS may be a client of the ORB. The ODBMS may want to use the ORB services such as location and naming (for distributed name services) or may use the ORB in order to access other ODBMSs, thus allowing heterogeneous ODBMS access. Current ODBMSs provide object identifiers that work only within one vendor’s prod-ucts, sometimes only within one database. The ORB object references could serve as a common denominator that allows selected object references and invocations in an ODBMS to span database boundaries (via encapsulating ODBMS object identifiers within ORB object references).
B.4 ODBMS as an Object Manager [top]
If every ODBMS object that an application wanted to reference were individually registered with the ORB or if every request to those objects in the ODBMS went through the ORB Basic Object Adaptor, the overhead would be unacceptable. This is equivalent to saying that every test of a bit of data or change of an integer must invoke the overhead of an RPC mechanism. Instead, the application should have the flexibility to choose which objects and which granularities are in fact known to the ORB, when re-quests to those objects go through the ORB, and be able to change this choice from time to time.
To achieve this maximum flexibility, we specify that the ODBMS has the capability to manage objects unknown or known to the ORB, to register subspaces of object identifiers with the ORB (to allow the ORB to handle requests to all of the objects in an ODBMS without the registration of each individual object), and to provide direct ac-cess to the objects it manages. For the objects unknown to the ORB, this direct access is provided via an ORB request to a containing object (e.g., a database), which then makes those objects directly available to the application. This provides consistency with and participation in the ORB environment and still provides the ODBMS with the ability to move objects around the distributed environment, cache them as appropriate, and provide efficient access.
For objects that the ODBMS has registered with the ORB, it may choose either to let requests to them execute the normal ORB mechanism or request from the ORB that any requests to those objects be passed to the ODBMS, perhaps for some period of time. Requests to such objects, whether through the ORB or directly to the ODBMS, must produce the same effect and be compatible with other users employing both mechanisms. In this way the ODBMS can provide consistency with the ORB and still coordinate with direct object requests.
The currently adopted OMG CORBA document provides for normal object access via the Basic Object Adaptor (BOA). For complete generality, flexibility, and interoperability, it executes via an interprocess mechanism (RPC for short) for every dispatch of every method. An extension is the Library Object Adaptor (LOA) that allows direct, considerably faster access to objects. After the first invocation (via the usual ORB mechanism), or through a compile-time optimization, a direct link is established to the object. Later access by the client to the object is then direct until the client notifies the ORB that it has released the object. We offer a new type of Object Adaptor, the Object Database Adaptor (ODA), to provide the ability to register a subspace of object identifiers and to allow access (including direct access) to the objects as if they had been individually registered.
The ODA provides a mechanism to register a subspace of object identifiers with the ORB rather than having to register all objects in the ODBMS. From the client’s point of view, the objects in the registered subspace appear just as any other ORB-accessible objects, with the same interface. The ODA should allow for the use of direct access (as in the LOA) to improve the performance of ORB/ODBMS applications.