Relational Database Extenders for Handling Complex Data Types

The foregoing copending applications are incorporated herein by reference.

A portion of the Disclosure of this patent document contains material which is subject to copyright protection. The copyright owner has no objection to the facsimile reproduction by anyone of the patent document or the patent disclosure, as it appears in the Patent and Trademark Office patent file or records, but otherwise reserves all copyright rights whatsoever.

1. Technical Field

This invention relates to relational database systems, and more specifically to a relational database system that handles complex data types emerging from advanced applications, such as multimedia, while maintaining application development productivity and reducing application complexity. The invention uses relational extenders to extend a database business table with new complex data types in that these new data types appear to application developers as seamless extensions to SQL (Structured Query Language).

For convenience, the following glossary of acronyms is provided for ease of understanding of the disclosure of this invention.

BLOB=Binary Large OBject

BMP=Bit MaP

CLOB=Character based Large OBject

DB2=Data Base 2

DBCLOB=Double Byte Character Large OBject

DBMS=DataBase Management System

GIF=Graphic Innterchange Format

GUI=Graphical User Interface

LAN=Local Area Network

LOB=Large OBject

MIDI=Musical Instrument Digital Interface

QBIC=Query By Image Content

RDB=Relational DataBase

SQL=Structured Query Language

TIFF=Text Image File Format

UDF=User Defined Function

UDT=User Defined Type

UMQ=Ultra Media Query

VOD=Video On Demand

2. Background Art

A. In General

Complex forms of data are emerging in many application domains, notably in multimedia (for example, image, video, and audio), but also in specialized application areas such as medical care (X-Rays, MRI imaging, EKG traces), geographical systems (maps, seismic data, satellite images), and finance (time series data). The new forms of data are sometimes referred to as "unstructured" data. In reality, the new data types are distinguished from traditional scalar numeric and character data by their multiple attributes, complex internal structure, and specialized behavior. For example, a video clip, as perceived by an application program, is not just a large binary field, it has attributes such as duration, format, number of frames, subject description, and information relating to legal ownership. It may also have associated index information, allowing search, positioning, and retrieval based on the content of individual frames. The video clip may have unique functions that could be used by an application program to retrieve selected portions of the clip, search for subject matter based on a combination of description and frame content, display, zoom, and edit the content, and return the clip to storage. These functions are clearly specific to the video clip data type and it would make no sense to perform them on some other data types, such as a static image.

Complex data types need to be stored in corporate databases, either in support of new applications, or to extend existing applications. In many cases, an application will require more than one new complex type in the same database or table. For example, a consolidated health care patient records system would probably require a database to contain many of the specialized medical data types mentioned above, as well as traditional data describing personal information, insurance details, etc.

In the database, complex data types must be capable of being searched, accessed and manipulated through the standard SQL language, without breaking the familiar table paradigm. Moreover, SQL support must be provided for the advanced technologies that are evolving to perform searches on the contents of large objects ("BLOBs"). Text search is a well known example, but algorithms also exist today for identifying images based on color, texture and shapes (for example, the Query By Image Content (QBIC) and Ultimedia Query (UMQ)), and the pattern matching techniques applied by IBM and Excalibur Technologies Corporation to fingerprint searching in the DB2.TM. Fingerprint Extender. Research is also proceeding into query by content of video and audio data. It must be possible to perform combined database queries that specify such content search criteria as well as criteria for traditional data search.

When the need for a complex data type is encountered in an application development project, a user might consider implementing the new type as an integrated part of the application. The attributes and structure of the data type might be stored in a special table; the handling of this table, and the special operations for the data type might be built into the application itself. Isochronous data such as video clips might be stored on a file server and the file names managed by the application program. However, such an approach has considerable disadvantages. The custom design and implementation of support for complex data every time it arises in an application would result in a large redundancy of effort and would add cost to advanced development projects. Also, the complexity of applications would increase because of the details of the data types that are visible in the application. Complex data types tend to require specialized skills for their definition and use, and by creating them as an integral part of an application, those skills will be needed in the application development team. The overall effect (and probably the most important disadvantage) is that a "roll-your-own" approach to handling these data types will ultimately slow the rate at which new advanced functions can be incorporated into corporate applications. The situation would be analogous to that which existed perhaps 30 or 40 years ago, when users tended to develop their own customized DBMS.

What is needed is a means of separating the specification and implementation of the structure and behavior of new complex data types from the development of the client applications that use them. An open architecture is required that allows the users themselves to create new data types, for use in many applications. The architecture should allow the installation of a new data type in a data base environment as needed, and independently of database product releases. It should mask the complexity of the data types from the applications. Effectively, it should bring complex data types under the umbrella of the data base system for all purposes of application access, security, and administration. In this way, user application development productivity may be improved, development complexity reduced, and the delivery of advanced function accelerated.

The challenge is to provide this open architecture that enables the development of specialized data types by users. The architecture must address the need to integrate the new data function seamlessly with existing RDB functions. The strategy must encourage the commercial development of new relational data types of broad applicability, both across industries, and within specialized industry domains.

This invention comprises Relational Extenders which are an architecture, a set of products designed to help RDB users handle emerging new complex data types in advanced applications and improve application development productivity, and reduce development complexity. Relational Extenders define and implement new complex data types in RDBs. The Relational Extender paradigm is essentially to extend corporate relational data with new data types'. Relational Extenders encapsulate the attributes, structure and behavior of new data types and store them in a column of a RDB table, such that they can be processed through the SQL language as natural additions to the standard set of RDB data types. Relational Extenders are separate from the database system itself, in the sense that they can be developed, installed and used independently of full database product releases, but they are also part of the database in that they appear to application developers as seamless extensions to SQL and the RDB product.

B. The Impact of Multimedia

Deployment of multimedia applications in the commercial environment is in its beginning stages. Today it is primarily deployed stand-alone on isolated work stations, but it is expanding to departmental LAN environments. In the 1994-1995 timeframe, customers in the commercial market are actively piloting departmental multimedia solutions with local or remote client access. As the LAN solutions mature, then, will evolve to enterprise LAN solutions in the 1996-1997 time frame.

Studies show that multimedia deployment in the commercial environment falls into five general application classes: (1) Presentations/Training/Kiosks, (2) Multimedia extensions to corporate data, (3) Query applications, (4) Commercial video on demand, and (5) Document management. Data management plays a role in all of these application areas because the multimedia data needs to be organized, preserved, managed, updated, searched and delivered to a potentially large number of users.

1. Presentations, Training, Kiosk:

This class encompasses applications which are normally developed with authoring tools. Some examples are: corporate dissemination of information to employees, travel agency type desktop sales demonstrations, and factory floor training modules. To date, the majority of multimedia applications are in this class and execute on stand-alone workstations; however, customers recognize the value of moving to client-server configurations for greater flexibility and functionality and to better leverage the use of multimedia.

2. Multimedia Extensions to Corporate Data:

This class consists of operational applications which utilize traditional corporate data in RDBs in combination with unstructured data (text, image, video, audio, etc.). These applications will initially be extensions of existing applications, but the ability to link corporate data with multimedia data will open up new application areas. One example of this application class would be a health care provider that wants to consolidate patient information, including image based test results and audio annotations by physicians, in a RDB. Another example is the addition of images or video clips to a retail catalog sales application for interactive shopping. Most commercial envirornments have traditional database applications today which can be extended with multimedia content in order to interact better with application users or consumers.

3. Query Applications:

Query applications are a subset of the other application classes. This class encompasses the case where the data being queried is a combination of traditional data and multimedia data. Examples include ad-hoc queries for decision support in retail buying or stock market analysis. But query may be used to determine what training modules are available or to find appropriate images to incorporate into a new application.

4. Commercial Video on Demand:

Video on demand (VOD) can be divided into two market categories: the commercial market and the consumer market. For purposes of this proposal, we are only addressing video on demand in the commercial market. In this market, end users access videos via a workstation at their desk, rather than a settop box and television. An example of commercial VOD is the dissemination of news clips to brokers in offices around a wide geographic area. Commercial VOD applications will be deployed in cases where the ability to deliver current information quickly is essential or where the quantity of information is so large that duplication of the information is impractical.

5. Document Management:

Document management applications are used in businesses where the requirement is to manage documents and records as images of the hardcopy originals and to group them by customer. A typical user is an insurance company which uses document management application to process policies and claims.

Data management plays an important role in the application areas described above. As collections of multimedia objects become large, efficient data management capabilities are required to organize, preserve, manage, update, search and deliver these objects to a potentially large number of users. So customers need ways of managing multimedia objects as business assets. Multimedia encompasses a wide range of data types which have not previously been associated with corporate data. Providing additional data types in the database makes it easier to relate multimedia data to traditional data. When multimedia data is managed by the database, customers can efficiently search on the attributes of the data, but new technology for searching the content of the multimedia objects can be deployed via the database too.

This invention is directed to relational extenders for a computer-based relational database. Each relational extender includes a first, business, table containing a user defined application database having at least one column dedicated to an object handle created by the user for defining the complex data type of one object; a second, attribute, table containing at least one column defining a unique characteristic associated with the one object and one column dedicated to containing the object handle; and a third, metadata, table containing at least one column defining a common characteristic associated with all objects defined within the business table and one column dedicated to containing the object handle and at least one column dedicated to containing a reference to object data associated with the object. The relational extender further includes a fourth table containing a reference to each object handle defined in the first table, a fifth table containing the names of the second and third tables for each object defined in the first table, a sixth table containing a reference to each object handle removed from the first table and including one column dedicated to containing a reference to the location of the object data associated with the removed object handle, and a seventh table having at least one column which describes a property of an object defined in the first table.

Additional features of this invention will become apparent from the following detailed description of the best mode for carrying out the invention and from appended claims.

FIG. 1 shows a high level overview of the database role in multimedia applications;

FIG. 2 shows the structure of the relational extenders of this invention;

FIG. 3 shows a general relational extender data model;

FIG. 4 shows a detailed relational extender data model;

FIG. 5 shows the data flow of a relational extender; and

FIG. 6 shows the interfaces between a relational extender and an application DBMS.

FIG. 7A-D shows multiple ways that text can be used in combination with other data types in a data table.

Relational extenders define and implement new complex data types and essentially extend relational data tables with these new data types. The relational extenders of this invention were designed to be implemented with DB2.TM. Version 2, available from IBM.TM. Corporation. It should be understood that the relational extenders of this invention are not limited to DB2.TM. Version 2. They may be implemented with any of a number of other RDBs. For convenience, however, in the following description of the best mode for carrying out the invention, the relational extenders will be described particularly in the context of their application to DB2.TM. Version 2.

FIG. 1 depicts the database role in multimedia applications. As noted above, studies show that multimedia deployment in the commercial environment falls into five general application classes: (1) Presentations/Training/Kiosks, (2) Multimedia extensions to corporate data, (3) Query applications, (4) Commercial video on demand, and (5) Document management. The application of interest to the customer is developed with application developer tools, which may include visualAge, MEDIA Script, IconAuthor, or Media Editors. The development tools utilize the infrastructure of a database system such as that of this invention of relational extenders, multimedia extensions (MCI, MMPM/2, Quicktime, client application enablers, and DB2). The database infrastructure access the underlying technology such as QBIC, text search, and UDFs and UDTs.

FIG. 2 shows the structure of the relational extenders of this invention, in which 202 represents a user application and tools. Various types of relational extenders 204a, . . . 204n, such as an audio extender, video extender, text extender, and user created extender, are logically associated with the object relational facilities 206 introduced by the RDB 208.

The foundation for the relational extenders is a set of object relational facilities introduced by DB2.TM. version 2. DB2.TM. is commercially available from IBM.TM. Corporation. The object relational facilities are an initial implementation of selected parts of the emerging SQL3 standard, and comprise user defined types (UDTs), user defined functions (UDFs), Large Objects (LOBs), triggers and checks.

A UDT allows a new data type, derived from a standard DB2.TM. data type, to be defined to the database engine. The database engine subsequently treats this UDT in a strongly typed manner, and ensures that an application uses a data item of a particular type only where such a data item is expected.

UDFs allow new functions to be defined, written in C or C++, or derived from built-in functions. Through the signature of its parameter list, a UDF may be associated with a particular UDT or standard type. A UDF may be used anywhere in an SQL expression that a standard DB2.TM. built-in function can use, and thus provide the mechanism for the seamless extension to the SQL language.

DB2.TM. triggers allow automatic dispatch of specific procedures whenever a relation table is the subject of an update, deletion, or insertion. Multiple triggers are allowed, with user control over their order of execution. DB2.TM. checks implement constraints on values stored in relation tables. A trigger or check is typically used to implement a business rule on a table, since it will always be invoked whenever the table is changed, without the knowledge or cooperation of the application code. In a relational extender context, triggers are typically used to maintain the internal structures or indexes of a complex data type.

Using these DB2.TM. building blocks, relational extenders treat each new and complex data type that is to be incorporated into a RDB as an object. Every object possesses hidden attributes that are accessed through a functional interface. An application program only has visibility to the interface, wherein the implementation details of accessing an object's attributes are invisible to the application.

FIG. 3 is a general relational extender model and FIG. 4 is a more detailed drawing of a relational extender data model. As shown in FIG. 3, an object has a unique object handle (identification label) 310 which the user application stores in a column of a business data table 312 to represent the object. The object handle 310 points to an entry in an Attribute Table 314. The entry in Table 314 in turn points to the location of an object BLOB in a hidden table 316 or in an external file 318. Alternatively, an object BLOB may be stored in a DB2.TM. column 311 of a business data table.

As shown in more detail in FIG. 4, a business table 312 column containing the object handle 310 is defined as the corresponding complex data type, or UDT. The interface to the object consists of UDFs, that take the object handle as a parameter and store, access, retrieve, search, and otherwise manipulate the object data.

An object may have multiple attributes which the database engine stores in separate tables 314. Those attributes that are common to all objects, regardless of the data type, are stored in a Base Metadata Table 412. Such attributes include: the name of the person creating the object, the date of creation, and the location of the actual object data. There is one Base Metadata Table for each user business data table. The table has one row 414 for each extender column 416 (a column that is enabled for a relational extender). Those attributes that are specific to the object data type are stored in the Attribute Metadata Table 418. There is one Attribute Metadata Table 418 for each enabled extender. For example, if a user business data table has been enabled for an image extender and an audio extender, it will have two Attribute Metadata Tables 418. An Attribute Metadata Table 418 may include such attributes for a video data type as the number of frames and the frame size, whereas for an audio data type, the table may include the attribute for the duration of play.

Alternatively, object attributes may be packed into the object handle 310 in the business table 312, depending on the size and number of attributes to be stored. For a small number of small attributes that are heavily accessed, this structure will provide faster access.

Some of the object attributes, either common or specific attributes, may be automatically set from the creation of the object. This would include static information contained in the header of the data, or information that the user provides upon creation. It may not be appropriate to change such information, in which case, the user application is only provided with functions to access these attributes, and not to set their values.

The object interface provides the user application with UDFs which access all of the attribute values, including the object data itself. Regardless of the implementation of object attributes, the object attributes are not directly accessible from a user application as columns in a relational table. These UDFs are strongly typed against their corresponding UDTs.

When the user application executes a function which alters the object data, a trigger may be initiated which will automatically update one or more object attributes related to the change. For instance, if a user application deletes a portion of a video object data, the number-of-frames attribute is automatically updated to reflect the new size of the video object data.

The choice of storage for object data is made when the user creates the object and imports it into the business data table. The object data may be stored either in a DB2.TM. BLOB column 311 of a business table 312 and directly accessible by the user application, or in a BLOB column 316 of a Base Metadata Table 412 or in an external file 318, both of which are under control of the object interface functions and thus hidden from direct access by the user application.

The invention also contains two or more relational tables 420 422 that are used for administration purposes. There is one Extender Information Table for each database engine. This table stores all of the relational extenders for which the database engine is enabled. It also records the UDT associated with each recorded extender. A Metadata Information Table also exists for each database engine. This table records the name of the Base Metadata Table and Attribute Metadata Table associated with a user table which has been enabled for a particular extender. There is a Delete Log Table for every extender which records all file names associated with extender columns in rows that are deleted. There is also a Column Information Table for each extender of the database engine. This table records the property characteristics of all the columns that are enabled for a particular extender. There is one row per extender column.

FIG. 5 shows the data flow of a relational extender. The user's application data table and object data are the bottom layer of the system. Metadata tables are maintained to manage and access the data tables. In addition, search support services (such as QBIC) are available for various object data types. Base extender services ensure data consistency management through the use of triggers that are specific for each object data type. Upon execution of a function that manipulates object data, the corresponding trigger automatically updates the object's metadata table and the user's application data. Other services enable application tables to define relational extenders, through metadata management, enable users to create unique functions with the use of SQL and UDFs, and provide users with a common set of UDFs including functions such as import. The next layer provides users with an interface containing functions that access object data and specific attributes of each relational extender and provide content search functions that manipulate the object data. Client functions exist for extender administration functions and tools such as allowing the user to set up data tasks and relational extenders. GUI functions provide the user with browse and playback capabilities of the object data. Content search preparation functions manipulate and search the object data based on its data type.

FIG. 6 shows the interfaces between a relational extender and an application DBMS. The user's application comprises a traditional business database table 312. The table contains at least one user-defined or predefined multimedia relational extender object 310. For each object defined in the data table, the object attributes are maintained in separate tables 418, 414. These tables include search functions appropriate for the corresponding object data type. The actual object data is maintained either internally 311, 316 or externally on a file server 318. The user application can accesses and manipulates the object data through the SQL application program interface (API). The SQL API in turn invokes DB2.TM. that access the object data directly. The DB2 interface includes standard DB2 functions, pre-defined multimedia relational extender functions, and UDFs. This structure hides the implementation details of object data from the user application. It provides the user with a fast and efficient mechanism of accessing and manipulating complex data objects.

Four extender products offer a public functional interface that is perceived as an extension to SQL, and all four interfaces share a similar look and feel, with common functions where appropriate. These extender products are: DB2.TM. Text Extender, DB2.TM. Image Extender, DB2.TM. Video Extender, and DB2.TM. Audio Extender. The data for all the data types can coexist in a table, and can be queried in the same SQL statement. DB2.TM. table resident data may be stored on either DB2.TM. for AIX version 2, or DB2.TM. for OS/2.RTM. version 2. External servers supported are the OS/2.RTM. Lan Server Ultimedia, and AIX.TM. RISC System/6000 platforms.

A. DB2.TM. Text Extender

The DB2.TM. Text Extender supports text documents stored in a business table as either DB2.TM. LOBs or as traditional DB2.TM. Character data types (VARCHAR, LONG VARCHAR). This choice of storage forms helps users to apply the text extender search technology to legacy text data that already exists in corporate DB2.TM. databases.

The DB2.TM. Text Extender provides UDFs that encapsulate full-text search technology as contained in the commercially available IBM.TM. Searchmanager product. The functions may be embedded in predicates of SQL queries in application programs, to search the content of text data contained in columns of a business table. SQL queries on a table may specify combinations of predicates on traditional data in the table, on text attributes (encoding format), and on text content in one or more text columns of the table at the same time, taking account of ranking information returned by the text search engine. Thus a single query may search on document date and author's name, which may be stored in standard data columns in the table, as well as on the content of the document, which may be stored in a LOB column.

The text search engine in the DB2.TM. Text Extender allows:

Synonym search, where an exact match of text does not need to be found. For example, a conceptual search on "airplanes" might result in matches not only on all documents about airplanes, but also on documents about helicopters and gliders.

Proximity search, where documents are chosen because they contain have words close to each other or within a sentence.

Boolean search, where documents are selected because they contain or don't contain certain words.

Wildcard searches, using front, middle, and end masking, as well as word and character masking.

The DB2.TM. Text Extender also supports searching hyphenated words and deploys linguistic search methods to find compound words. The response to a search is a list of documents that satisfy the conditions.

The DB2.TM. Text Extender supports loading documents into the DB2.TM. database from a variety of industry-standard document formats. Linguistic functions (such as synonyms) are available based on dictionaries for 17 national languages. Linguistic search is supported for that subset of document formats that identify the languages of their content. Search support data (indexes) are initially created from data loaded into the DB2.TM. table through supplies data preparation facilities. The indexes are updated from data collected automatically by triggers when application programs maintain the documents in the table.

Overall, the DB2.TM. Text Extender provides a UDT to represent the index column type in the business table and UDFs to search and retrieve selected text attributes (for example, text format). Triggers are provided for purposes such as maintaining a log of changes to the business table, or periodic updating of the index tables. Administrative facilities are provided for establishing and maintaining indexes. A highlighting text browser function (or GUI component) is included which users may invoke from within their application programs to display retrieved documents and to show the "hit" words or phrases.

Text search is a fundamental search technology, and may be applied to the retrieval of many other kinds of data. Most kinds of data can be described by keywords or by an abstract, and text search can be used against these descriptions to locate the subject data item. FIG. 7 shows some of the many ways that text can be used in combination with other kinds of data in a table.

In interpreting the scenarios in FIG. 7, it should be noted that the DB2.TM. Text Extender stores its subject text document in a column in the business table, and introduces a second column for indexing purposes. The second column contains the UDT which "owns" the search UDFs.

FIG. 7(a) An existing text string 714 is stored in a column 712 of a business table. The column type may be VARCHAR, LONG VARCHAR, or CLOB (character based LOB). The additional column 710 is the indexing view of the text, hiding the text search engine 708 along with its search support data (indexes). The programmer deals with this column for search operations.

FIG. 7(b) Here a document 716 is stored as a LOB. Again, the document is indexed by the search engine 708.

FIG. 7(c) Here the indexed text describes an image 718 stored in another column. The image might be a simple BLOB, stored as a DB2.TM. LOB data type, or it might be a DB2.TM. Image Extender type, stored in a hidden location, and with its own public functional interface of UDFs. The text 714 column (plus its index column 710) provide a means of locating desired images by their text description.

FIG. 7(d) Here, a column in the application table contains a reference 720 to a file stored externally on a server (in the form of a path and filename). The file is considered to be under the control of the application (i.e., not managed by any extender) and might contain any kind of data relevant to the application. This is a typical approach to handling "unstructured" data in today's data base environments. But here, too, the Text Extender may be used to index that file data, through a column containing a textual descriptive information 714 and its associated index column 710.

Note that all these scenarios share a common characteristic, that in each case the Text Extender might be retrofitted to an existing business table. The character column in case FIG. 7(a), the character LOB in case FIG. 7(b), the image column in FIG. 7(c), and the external file reference in FIG. 7(d) might all have existed in the business table prior to the addition of the Text Extender indexing column. And, after addition of the Text Extender, each of these data items will continue to be available unchanged to its existing applications.

In all these cases, also, after addition of the Text Extender, combined searches of business data in the table, and text content are possible in a single SQL statement. For example, an SQL statement might request all rows that have keywords X and Y in the same paragraph of the text column, and where the amount is less than 100 in the price column.

B. DB2.TM. Image, Video, and Audio Extenders

In addition to the Text Extender, three DB2.TM. multimedia extenders provide the functionality for each of the three primary multimedia data types: image, video and audio.

Since it is not a high priority for these multimedia extenders to support legacy data in existing tables, all support the storage of their BLOB data in either:

The database, in tables controlled by the extender. This data may be resident on DB2.TM. servers running either DB2 for AIX.TM. V2 or DB2.TM. for OS/2.RTM. V2.

External file servers for OS/2.RTM. and AIX.TM..

UDFs are provided to create new instances of the multimedia data types, and to return a handle which the application program stores in the business table. These UDFs build all the underlying structure of the instance, and store the multimedia BLOB data in its chosen location. The functions will accept input data for the multimedia instance from program buffers, or from a file. If the multimedia data is to be stored on an external server, it is possible to adopt a file already stored on the server without moving it from its existing location.

UDFs are also provided to retrieve information from a multimedia type column in a table, including the descriptive attributes of the type (for example, image size), and the BLOB data. The data access functions hide the physical location of the BLOB data from the application. Data stored in the database is accessible only through the access functions. Data stored on an external server may also be processed through the access functions, or alternatively the application may obtain the physical path of the data file on the server and use it to process the file directly, or stream it through some suitable service such as MMPM/2.

A variety of multimedia formats are supported for each type (such as TIF, GIF, BMP for image and WAVE, MIDI for audio). In many cases, the extender extracts attributes from the headers of these multimedia objects as they are loaded into the relational table, relieving the programmer of the need to supply them explicitly. The extenders also accept unrecognized formats for storage and offer the programmer the option of specifying the attributes. In selected cases, conversion routines are provided to convert data between formats on input and output.

Each of these DB2.TM. Extenders comprises the UDT defining its data type, the UDFs providing its behavior, and triggers to maintain integrity of the extender data structures as business tables containing extender data columns are copied, updated, and deleted. GUI components are provided for certain purposes. For example, rendering routines are included for each extender, and for each supported client platform. Users can optionally incorporate these routines into their applications. Administrative functions are provided for preparing a table to contain a column for an extender data type, and for other housekeeping purposes. A common installation utility is provided for installing the extenders.

This section describes how MultiMedia (MM) Objects are associated with business data, and where the attributes and the BLOB data of an object may be stored. The section also provides design details about data models, rationales, and scenarios. The discussion is centered around an 20O Image Extender; however, the information presented is applicable to other extenders, such as Audio and Video Extender. Sections follow describing the core MM Application Programming Interface (MMAPI) functions.

A. Scenario and SQL Design

Assume that the user created a multimedia personnel database. Within the personnel database, User.sub.-- A created Table.sub.-- A with these statements:

User.sub.-- B created Table.sub.-- B with these statements:

Note that DB2Image is a User Defined Type (UDT) created by:

CREATE DISTINCT TYPE DB2Image AS VARCHAR(254) WITH COMPARISONS;

1. Data Model

Within a database, all extenders (image, audio, video) share a special qualifier, mmdbsys, all metadata tables and triggers are created and owned by this special qualifier.

When a database is first enabled (for any extender), the base layer creates 2 administrative support tables:

mmdbsys.ExtenderInfo

mmdbsys.MetatableNames

For each extender, there are 4 additional administrative support tables:

mmdbsys.xxxColumns

mmdbsys.xxxBaseImport

mmdbsys.xxxAttrImport

mmdbsys.xxxDeleteLog

Where `xxx` is up to the first 8 characters of the extender name. For example, administrative support tables for Image Extender are named: ImageColumns, ImageBaseImport, ImageAttrImport, ImageDeleteLog.

For each user table that is enabled for an extender, 2 corresponding metadata tables are created: xxxB+time() and xxxA+time(), where `xxx` is the extender name and time() is the current time expressed in number of elapsed seconds since 12:00am Jan. 1, 1970. For example, ImageB12345678 and ImageA12345678. The ImageB12345678 table holds metadata that are similar for all extenders (e.g.: filename, size, remarks, etc.). The ImageA12345678 table holds metadata specific for Image Extender (e.g.: width, height, number of colors, etc.).

For each image in the user table, there is 1 row in each of the 2 image metadata tables.

There is no limit on the number of image columns a user can have on a table.

The user can choose to store images in DB2 BLOB or in file system.

Within a database, all handle values are unique, persistent tokens. The handle value serves as a key to identify an image.

2. Handle Encoding

Handle value is encoded as follow:

The ":" character is used as the delimiter.

3. Scenario

The Image Relational Extender design is best illustrated using a given scenario. Assume that it is desired to create the personnel database mentioned earlier. This section explains events that happen at each of the steps (create database, create table, alter table, insert, select, update, delete).

a. Create Database

User Does:

Note that caller of DBiEnableDatabase() must have DBA or equivalent authority. Also, the application must establish a valid connection to the database prior to calling this API.

DB2/CS Does:

Create the database.

Create the system tables

etc.

Image Extender Does:

Inside DBiEnableDatabase():

Note that the UDT, all metadata tables and UDFs are created under mmdbsys userid.

b. Create Table

Now User.sub.-- A wants to create Table.sub.-- A.

User Does:

Note that caller of these APIs must have "alter table" or higher authority on the subject table. Also, the application must establish a valid connection to the database prior to calling these APIs.

MMDM Does:

Inside DBiEnableTable():

Inside DBiEnable Column():

Image Extender Does:

Inside DBiEnableColumn(), working together with.sub.-- bestENableColumn():

A new row is added to mmdbsys.ImageColumns:

tQualifier="User.sub.-- A"

tName="Table.sub.-- A"

usercolumn="picture"

basemetatableName="mmdbsys.ImageB12345678

attrMetatableName="mmdbsys.ImageA12345678

storageType=DBI.sub.-- EXTERNAL

controlSource=`N`

deleteSource=`Y`

Create the insert, update, delete triggers on User.sub.-- A.Table.sub.-- A just like those in section"2.3.2 Create Table", substituting "mgrPicture" for "picture."

d. Insert with VALUES Clause

Suppose User.sub.-- A want to add a record for Tri Ha into Table.sub.-- A.

User Does:

Image Extender Does:

Inside the first DB2Image():

Since it is not known what user table or what user column this image is for, a temporary row is added to the ImageBaseImport table. The "insert triggers" will move it to the correct metadata table later. Depending on what version of DB2Image() UDF is called, the storage type for this image can be deduced. A unique, persistent handle is returned which serves as the key to identify this object.

Another temporary row is added to the IamgeAttrImport table. At this time, automatic processing is done to compute all image specific metadata (number of colors, format, size, thumbnail, etc.). These metadata (along with user marks) can be used later for queries.

Inside the second DB2Image():

A new row is added to the ImageBaseImport table:

Another new row is added to the ImageAttrImport table:

Once this is completed, two insert triggers (one for picture, one for mgrPicture) are fired.

e. Insert with Full Select

Suppose that Table.sub.-- A had been fully populated. User.sub.-- B created a test table, TB.sub.-- test, with the same schema as Table.sub.-- A. He also called DBiEnableTable() and DBiEnableColumn(). Two metadata tables, mmdbsys.ImageB87654321 and mmdbsys.ImageA87654321 were created for TB.sub.-- test.

Now User.sub.-- B wants to fill TB.sub.-- test using data stored in User.sub.-- A.Table.sub.-- A.

User Does:

Image Extender Does:

The SQL statement above would cause the "insert triggers" defined on TB.sub.-- test to fire.

Note that there are now 2 copies of the same BLOB, one is User.sub.-- A.Table.sub.-- A.picture, another in no User.sub.-- B.TB.sub.-- test.picture.

Also note that even though User.sub.-- A.Table.sub.-- A.mgrPicture and User.sub.-- B.TB.sub.-- test.mgrPicture point to the same file, the handle value in Table.sub.-- A.mgrPicture is not the same as the handle value in TB.sub.-- test.mgrPicture because the table name is encoded in the handle.

f. Select

Suppose User.sub.-- B wants to see the thumbnail images of all employees whose pictures were taken sometime in 1994.

User Does:

Image Extender Does:

For each row of TB-test, the DBMS invokes Remarks(), which returns the user remarks on that image. If any of them match the search pattern in the WHERE clause, DBMS invokes Thumbnail(), which returns bit stream of type LONG VARCHAR (sizes vary from 3K bytes to 9K bytes). Note that the user is still responsible for displaying the thumbnail using whatever operation system supports that he/she has.

g. Update

Design note Feb. 2, 1995: update comes in 3 flavors:

These two cases are treated as get/set object attribute, thus the handle value does not change. The third case involves changing both image content and user remarks, this is treated as creating a new object. The only way a user can do this is by calling DB2Image(). In this case, a new handle value is returned.

Suppose User.sub.-- B wants to update a person's ("NAME") picture and its associated remarks.

User Does:

Another new row is added to the mmdsys.ImageAttrlmport:

The old rows in the metadata tables are not changed. Note that the BLOB or the old metadata rows cannot be deleted because the current SQL statement may be rolled back. However, if a new row is added successfully, the triggers are fired. At this point, the old metadata rows are deleted. The external image files are only deleted by calling DBiReorgMetadata Tables( ).

h. Delete

User.sub.-- B wants to delete NAME's record from the TB.sub.-- test table.

User Does:

Image Extender Does:

The "delete triggers" are fired.

B. Metadata Table and Trigger Descriptions

This section describes the table definition and table naming convention for all tables created by the extenders. It will also document the naming convention for triggers and handles.

Following is a summary of all the tables created and used by the extenders:

1. Multimedia Column (MM Column)

A column is called MM Column if it is enabled for a multimedia extender.

2. Extender Information Table (one per database)

There is only one such table for each database. It is used to record all the multimedia extenders that this database is enabled for. It also records the UDT associated with the extender.

a. Naming Convention

MMDBSYS.EXTENDERINFO

b. Table Definition

3. Metadata Information Table (one per database)

There is only one such table for each database. It is used to record the name of base metadata table and attribute metadata table that are associated with a user table which has been enabled for a particular extender.

a. Naming Convention

MMDBSYS.METATABLENAMES

b. Table Definition

4. Delete Log Table (one per extender)

This table is to keep track of all the file names associated with MM columns in rows that are deleted.

a. Naming Convention

MMDBSYS.xxxDELETELOG

where xxx is the up to 8 character extender name.

Example: MMDBSYS.IMAGEDELETELOG is the delete log table name for IMAGE extender.

b. Table Definition

where xxx is the up to 18 char UDT associated with the extender.

Example, delete log table definition for Image extender with UDT DB2Image will be

5. Column Information Table (one per extender)

This table is to record the property of all the MM columns that are enabled for a particular extender. There is one row per MM column.

a. Naming Convention

MMDBSYS.xxxCOLUMNS

where xxx is the up to 8 character extender name.

For example, column information table name for image extender is MMDBSYS.IMAGEDELETELOG.

b. Table Definition

6. Base Metadata Table (one per user table and extender) and Import Base Metadata Table (one per extender)

There is one base metadata table for each user table and extender. If a user table has been enabled for image extender and audio extender, it will have 2 base metadata tables (one for image extender and one for audio extender). If a user table has been eneabled for image extender only, it will have only one base metadata table.

Base metadata table is used to store information that is common to all extenders such as importer, import time, filename, size, . . . , etc.

The base metadata table will have one row for each MM column in a row in the user table. For example, a user table is enabled for image extender and the user table has 10 MM columns. As rows being inserted to the user column, rows are also inserted into base metadata table. After successfully inserting 5 rows to user table, 50 rows will be inserted to the image base metadata table associated with this user table. The MM column value is used as a key to the base metadata table to retrieve all the related metadata.

The rows in the base metadata table are inserted by the MM extender's insert trigger created for the MM column. The rows in the base metadata table are updated by the MM column's update UDF. The rows in the base metadata table are deleted by the MM columns' delete trigger.

It would be ideal if the rows in the base metadata table can be inserted by the extender import UDF. However, the import UDF knows about database name and extender name but not the user table name. So the import UDF will insert rows to the import base metadata table. Since the after insert trigger know about the user table name, so it will then transfer the rows from import base metadata to the base metadata table associated with the user table.

For this reason, the table definition for these 2 tables are identical.

a. Naming Convention

i. Import Base Metadata Table (one per extender)

MMDBSYS.xxxBASEIMPORT

where xxx is up to 8 character extender name

For example, import base metadata table name for IMAGE extender MMDBSYS.IMAGEBASEIMPORT.

ii. Base Metadata Table (one per user table and extender)

MMDBSYS.xxxBttt

where xxx is up to 8 character extender name and

ttt is 9 character time stamp in seconds since 1980

For example, MMDBSYS.ImageB12345678 is a valid base metadata table name for Image extender.

b. Table Definition

For example, xxx will be sent to DB2Audio if this table is defined as the base metadata table for Audio extender with UDT defined as DB2Audio.

7. Attribute Metadata Table (one per user table and extender) and Import Attribute Metadata Table (one per extender)

These tables are used to store extender specific information for each MM column. Again there is one row in the attribute metadata table for each MM column in a row in a user table.

The MM column value is used as a key to the attribute metadata table to retrieve extender specific metadata. The reason for the existence of the import attribute metadata table is the same as that for import base metadata table.

a. Naming Convention

i. Import Attribute Metadata Table (one per extender)

MMDBSYS.xxxATTRIMPORT

where xxx is the up to 8 character extender name

For example, import attribute metadata table name for IMAGE extender is MMDBSYS.IMAGEATTRIMPORT.

ii. Attribute Metadata Table (one per user table and extender)

MMDBSYS.xxxAttt

where xxx is the up to 8 character extender name and

ttt is 09 character time stamp in seconds since 1980

For example, MMDBSYS.ImageA123456789 is a valid attribute metadata table name for Image extender.

b. Table Definition

The minimum requirement for this table is to have the HANDLE column. It's extender developer's responsibility to complete this table definition.

HANDLE xxx NOT NULL PRIMARY KEY

where xxx is the UDT defined for the extender

For example, xxx will be set to DB2Video if this table is defined as an attribute metadata table for Video extender with UDT set to DB2Video.

8. HANDLE column in Metadata table

The handle in both base and attribute metadata tables holds the same values as the associated MM column in a user table. Most of the extender's UDFs take handle as one of the input argument. The handle values follows the following format so that UDF can take advantage of it. The handle is always a string of length 111 for now even though it's base type is indirectly defined as VARCHAR(254) in the metadata table.

9. Trigger Naming Convention

The only requirement is that the all upper case extender name should be the first 8 characters and then follow by a time stamp. The trigger name has extender name encoded so that we can disable a particular extender for a user table by dropping all the triggers associated with that user table and extender.

10. Trigger Definition

Triggers are defined for each MM column. All triggers work in the same way regardless of which database, which user table, which column in the table, or even which extender. To make it easier to understand, all triggers in this section are documented using MM column PICTURE in user table "USERID.TABLENAME" within database `DATABASE` for IMAGE extender with UDT set to DB1Image. The base metadata table is `MMDBSYS.IMAGEB123456789` and the attribute metadata table is `MMDBSYS.IMAGEA124356789`.

The extender, user table, and MM column dependent information in the trigger body is underlined and in italic style. The.sub.-- bestEnableColumn() API will create these triggers.

e. After Delete Trigger froM MM COlumn with DeleteSource Set to `N`

f. After Delete Trigger for MM COlumn with DeleteSource set to `Y`

C. Data Model

MM objects consist of MM BLOBs that are the actual data of the text, image, etc., and MM Attributes that relate to the BLOB, such as "date of creation". MM objects are identified by a unique identifier called a handle, and business tables reference MM Objects through columns that contain these handles. The handle is the actual UDT defined to the database engine as the MM Object data type, and so a separate column in the business tables is required to represent each MM Object type.

Non-BLOB MM Attributes are accessed through UDF's only. Some attributes, such as "date of creation", are common to all MM Object types, while other attributes, such as "number of frames", are expected to be dependent upon the particular type of MM Object. Furthermore, some MM Attributes may be system generated, e.g., date of creation, or automatically inferred from the contents of the BLOB or its header, while others will need to be specified when an MM Object is created.

MM BLOBs may be stored in the business table in the database, or in a hidden table, or in an external server. The MM BLOBs (regardless of their location) may be accessed in the same manner as other MM Attributes, namely through the handle of the object via UDF's. If this is done, the application can be relatively insensitive to the physical location of the BLOB. BLOBs stored in the business table may also be accessed directly by applications. If this option is chosen, the application will be sensitive to the location of the BLOB.

Streaming of MM BLOBs is accomplished by obtaining the access path (an MM Attribute) of an externally stored MM BLOB, and passing this filename to a service such as MMPM/2 that can handle streaming.

D. Data Access Functions

db2xxxxx() is the constructor function that creates new instances of MM Objects of type xxxxx. There is at least one different constructor function for each MM Object type, for example, db2Image() and db2Audio(). db2xxxxx() returns the value of an object handle of the appropriate type. To accommodate alternative combinations of optional parameters, there may be a family of similarly named, overloaded constructors for each MM data type.

Source may be a DB2.TM. LOB type host variable, containing the incoming multimedia data, that may be represented by a LOB type buffer, a client file reference, or a LOB token. The data will be moved to a specified destination.

Source may also specify, through a fully qualified path, a reference to a file that is accessible to the multimedia object server. The file will be "adopted" without copying it from its existing location.

{Dest} may be an optional indicator of where the BLOB is to be stored, either in the business table, in a hidden table, or an external server. If an external server is indicated then a destination file path may be specified. The BLOB data will be copied from the source to the {dest} (unless the source file is to be "adopted" in its present location). If the parameter is absent, storage will be to a system wide default destination, either to a hidden table, or a directory path on an external server.

{Source.sub.-- format} and {Dest.sub.-- format} may optionally be specified. {Source.sub.-- Format} indicates the format of the incoming BLOB. (Note that the source format may be implicitly defined by the BLOB itself). {Dest.sub.-- format} indicates the format of the BLOB as it is to be stored. If the formats are different the BLOB will be converted before it is stored. The extender should tolerate the absence of these parameters, basically by storing BLOBs as-is.

{Index} may be an optional parameter(s) that controls indexing actions, if supported by the extender (for example, specification of which index to put this particular object in). System-wide defaults will apply if the parameter is absent.

{Attr1, Attr2, . . . } may be a list of optional attribute values for this data type. Absent parameters will cause a suitable default value to be stored. Note that the values of some MM attributes may be generated automatically by the constructor.

The content() function provides applications with access to MM BLOBs, content(handle) takes a handle and returns a LOB data type, which may be fetched into a LOB host variable (buffer, client file type, or a LOB token).

This simple form of the function is overloaded for all multimedia data types. There may be other forms of content() that are specific to particular data types, and include parameters specific to that type (option1, option2, etc.).

The content(handle, source) form of the function updates the MM BLOB value of an object represented by handle (and any other attributes that are automatically inferred from the BLOB or its header). It is overloaded for all data types.

This function returns a char variable containing the full pathname of BLOBs that are stored in external servers, or table and column names of BLOBs that are stored in the database. It is overloaded for all data types.

The attribute.sub.-- name(handle) functions are names for each attribute of the data type, and return the value of that MM attribute. For example, there may be DateofCreation, AttributeOwner, etc. Some functions will be MM Object type dependent (e.g., NumberVideoFrames()), and others will be type independent (e.g., DateofCreation). The latter functions should be overloaded for all data types.

The attribute.sub.-- name(handle, value) form of the attribute functions will update the value of the named attribute of an existing MM Object. For example, owner(handle, "Don") changes the attribute "owner" of the object represented by "handle" to "Don".

E. Content Search Functions

contains (col.sub.-- name, criteria)

col.sub.-- name identifies the objects to be searched.

Criteria are the (type dependent) search criteria.

The contains() function is overloaded for all data types. The content of Criteria parameter will differ between types, but is expected to take a common form that will be returned by all the GUI buildPredicatexxxx() functions (see below). This will allow the applications to set up searches on tables containing various data types without needing to use a different form of predicate for each data type.

The contains() function returns a value that indicates how well the search criteria have been met. The form of the function will allow its use as a predicate in a WHERE clause, and/or as the subject of an ORDER BY clause to rank returned rows. Value is between 0 and 100.

Note that search engines may, in addition, provide type dependent content search functions for more specialized forms of search.

F. Content Functions

The list of Content Functions is data type dependent.

Some examples might be:

(1) Content manipulation, e.g., rotate().

(2) Network traffic reduction, e.g., compress().

(3) Format transformation, e.g., TIFF.sub.-- to.sub.-- GIF.

(4) Other, e.g., encrypt.

(Note that format transformation might be a candidate for a content function, as well as being a "built-in" facility of the Import and Export functions described above. There may be other functions, as well as these conversion functions, where there is value in providing the functionality in both forms.)

G. Support for SQL Operations

All extenders are expected to handle the following forms of SQL operations on a subject business table:

1. Setting Extender data type column to NULL

When the Extender data type column value is set to NULL, the data type instance should be deleted, in the sense that it is subsequently inaccessible by applications. Actual physical clean up is implementation dependent.

2. Row delete

When a row of the business table is deleted, an extender data type instance in that row should be deleted.

3. Table drop

When a table containing Extender data type columns is dropped, all extender data type instances in that table should be deleted.

4. Copies

When an Extender data type is copied into another table location:

a. A new logical copy of that data type instance should be created, in the sense that the original copy and the new copy will appear to applications as distinct instances, and all subsequent operations (update, delete, etc) to these two instances will be independent. (Whether the data is actually physically copied, and when, is implementation-dependent.)

b. The data type instance that was previously held in the copy destination (if any) should be deleted.

H. Client GUI Functions

GUI "functions" are assumed to be in the form appropriate for the main programming environments on each client platforms.

There are two GUI functions that are considered core functions, and should be provided by all extenders.

This function will open a suitable basic dialog on the screen and elicit information from the user on the search criteria for the data type xxxxx. It will update a criteria variable that can be embedded in a content search "contains()" function in the WHERE clause of a SQL SELECT statement. The content of the criteria will differ between data types, but the declaration of the criteria should be common for all data types.

Note that extenders may, in addition, provide type dependent GUI components that establish specialized dialogs with the user to perform specialized forms of content search.

The renderxxxxx() function will take an object of type xxxxx and render it on an output device in an appropriate way.

While various embodiments of the present invention have been described above, it should be understood that they have been presented by way of example, and not limitation. Thus the breadth and scope of the present invention should not be limited by any of the above described exemplary embodiments, but should be defined only in accordance with the following claims and their equivalents. It will be understood by those skilled in the art that various changes in form and detail may be made therein without departing from the spirit and scope of the invention.