Henrike BertholdDRESDEN UNIVERSITY OF TECHNOLOGY
DATABASES
Schema design and query processing in a federated multimedia database system
CoopIS, Trento, Italy, 05/09/2001
Problem
A Federated Multimedia Database System
Global schema construction
Query processing
Summary
Henrike Berthold & Klaus Meyer-Wegener
Dresden University of Technology, Germany
Henrike BertholdDRESDEN UNIVERSITY OF TECHNOLOGY
DATABASES
Problem
• SDBS: Database system with a schema, a query language, ...• MRS: Media retrieval system• MSS: Media storage system
Application 1 Application n
SDBS 1 SDBS k MRS 1 MRS m MSS 1 MSS n
Henrike BertholdDRESDEN UNIVERSITY OF TECHNOLOGY
DATABASESMedia retrieval systems• Perform content-based search based on automatically extracted
features (perception + semantical primitives)• work for a set of media objects of a certain media type such as
image, text, video, audio or a subset with a certain semantics such as passport photos
• Retrieval functions– single: search argument (example, feature, weighting),
media object -> similarity value– coll: search argument -> list if tuples (similarity value, MID)– result restrictions of coll: lower bound for similarity values,
subset of media objects, top n• Example systems: QBIC (IBM), Excalibur Image Datablade,
Melodiscov (LIP6)• Problem: only interactive interface and no programming one
Henrike BertholdDRESDEN UNIVERSITY OF TECHNOLOGY
DATABASES
Media storage systems
• Store and manage media objects like images, videos• provide operations for those• support especially timed playout of continuous media objects• all systems provide media description data such as size, format• some systems support transactional concepts like durability,
atomicity• Example systems: Tiger Shark (IBM), Symphony (Uni Texas),
Fellini (Bell Labs), Kangaroo/Memo.REAL (Dresden University of Technology)
Henrike BertholdDRESDEN UNIVERSITY OF TECHNOLOGY
DATABASESA Federated Multimedia Database System
- Database schema and query processing
- Operations on media objects
- Transactions
FMDBS
Application 1 Application n
SDBS 1 SDBS k MRS 1 MRS m MSS 1 MSS n
FMDBMS
Henrike BertholdDRESDEN UNIVERSITY OF TECHNOLOGY
DATABASESGlobal schema construction
• Structured-data schema (SDS) = global schema of all SDBS-Schemas
MSS1-Schema MSSn-Schema
Media Schema (MS)
SDBS1-Schema SDBSk-Schema
Structured-data Schema (SDS)
integrates
based on two intermediate schemas:
• Media schema (MS) = global schema of all MSS schemas
Henrike BertholdDRESDEN UNIVERSITY OF TECHNOLOGY
DATABASESExample media schemaCommon data model: ODMG, CROQUE• Types and subtype-relation• Classes and subclass-relation
Image:TImage
CityPart:TImagePartPassportPhoto:TImage City:TImageWithPart1 *
Video:TVideo Audio:TAudio Text:TText
ContMedia:TContMedia
:TSingleMedia :TObject
Henrike BertholdDRESDEN UNIVERSITY OF TECHNOLOGY
DATABASESStep 1: Base integration
Schema architecture:
MSS1-Schema MSSn-Schema
Media Schema (MS)
SDBS1-Schema SDBSk-Schema
Structured-data Schema (SDS)
integrates
Global Schema
Henrike BertholdDRESDEN UNIVERSITY OF TECHNOLOGY
DATABASESStep 2: Integration of media retrieval functions
• single: objekt method +• coll: class method -, function -
• allows combination of boolean queries with media retrieval queries
Henrike BertholdDRESDEN UNIVERSITY OF TECHNOLOGY
DATABASES
Step 2: Example
t5 (=t1‘)
t6 (=t3‘) t7 (=t4‘)
c2:t1 c3:t2c1:t1
c6:t4 c7:t3c5:t3c4:t3t2
t0
t1
t3 t4
t0t0
t1
t3
t0
t4
t0
t1
t3
t0
c11:t6 c12:t6 c13:t7
c1:t5
Henrike BertholdDRESDEN UNIVERSITY OF TECHNOLOGY
DATABASESStep 3: Extension
• complex extension: new objects + relationships
TVideoTText
TSingleMediaTLecture TSlide
TImage
1 1..*TVisualElement
Time interval Region
TVideoElementTTextElement
TImageElement
2 8x
y
Region forvideo
Region forimage
Region for text
t in min
Text1
Image1
Video
Image2
Text2
• Simple extension: relationships
Henrike BertholdDRESDEN UNIVERSITY OF TECHNOLOGY
DATABASES
Schema architecture
MSS1-Schema MSSn-Schema
Media Schema
SDBS1-Schema SDBSk-Schema
Structured-data Schema
integrates
Global Schema
Internal Schema
MRS1-Function MRSm-Function
Henrike BertholdDRESDEN UNIVERSITY OF TECHNOLOGY
DATABASESConstruction data
Henrike BertholdDRESDEN UNIVERSITY OF TECHNOLOGY
DATABASESQuery processingGlobal OQL Query
Global querycomputation
Decomposition andglobal optimization
Modification
Local optimizationand execution
Global execution
Result
Type checking andmapping to intermediate representation
Replacing global specifiers by local ones
Decomposition and optimization by theglobal query processor
Optimization by the local queryprocessors and local execution
Combining local results globally
Constructiondata
Henrike BertholdDRESDEN UNIVERSITY OF TECHNOLOGY
DATABASES
FMDBS-specific problems
• Modification– Vertical fragmentation
GT1 GT2GT3
GT4
complete replacement of a global object is not possible
– Replacement of search methods
Henrike BertholdDRESDEN UNIVERSITY OF TECHNOLOGY
DATABASESComprehension syntax
Henrike BertholdDRESDEN UNIVERSITY OF TECHNOLOGY
DATABASESMonads and algebras
Henrike BertholdDRESDEN UNIVERSITY OF TECHNOLOGY
DATABASES
Advantages
• uniform representation of collections (set, bag, list) and aggregations
• readability• there are equivalences defined
– easy manipulation
• there is a mapping to an algebraic expression– embedding in traditional query processing is possible
• thus: better suited for manipulations than an object algebra
Henrike BertholdDRESDEN UNIVERSITY OF TECHNOLOGY
DATABASESModification
Henrike BertholdDRESDEN UNIVERSITY OF TECHNOLOGY
DATABASES
Improvements
• Paths that belong to one source are replaced completely; so far: replacement of each single function
• A generator which binds a global object is replaced by a sequence of qualifiers which produce all local objects that are used in the query; so far: production of local objects there, where they are used
• exploit function coll and its result restrictions
Henrike BertholdDRESDEN UNIVERSITY OF TECHNOLOGY
DATABASESReplacement of function single by coll
Henrike BertholdDRESDEN UNIVERSITY OF TECHNOLOGY
DATABASESIdentification of a lower bound
Henrike BertholdDRESDEN UNIVERSITY OF TECHNOLOGY
DATABASESIdentification of a subcollection restriction
Henrike BertholdDRESDEN UNIVERSITY OF TECHNOLOGY
DATABASESPrototype
Henrike BertholdDRESDEN UNIVERSITY OF TECHNOLOGY
DATABASES
Summary• An FMDBS manages structured data and media
data. It takes the integration efforts from the applications.
• Development of a procedure to construct the global schema– data (relationships, new types and classes) and media
retrieval functions can be added
• Development of a procedure to modify queries– can handle the vertical fragmentation– choses efficiently executable media search functions
• Applications have a powerful mean to find suitable data
Henrike BertholdDRESDEN UNIVERSITY OF TECHNOLOGY
DATABASES
Outlook
• Problem: Chose a result restriction if some are applicable
• Need of a general cost model of media search systems
• Construction of a complete FMDBS• Need of tools which support the construction
and the administration