Data Modeling using XML. Murali Mani, WPI Antonio Badia, University of Louisville Oct 13, 2003

Transcription

1 Data Modeling using XML Murali Mani, WPI Antonio Badia, University of Louisville

2 Outline Part I: How to come up with good XML designs for real world database applications? Part II: Translation between Relational and XML models.

3 Part I: How to come up with good XML designs for real world database applications?

4 What is XML? <book> <author>j. E. Hopcroft</author> <author>j. D. Ullman</author> <publisher name= Addison-Wesley /> </book> book author author publisher J. E. Hopcroft J. D. Ullman name= "Addison- Wesley"

5 XML for information exchange

6 XML Publishing Text applications Database applications X gave two <person> <reviewer>x</reviewer> thumbs up rating <Name>A</name> Age Salary gave <rating>two thumbs to the movie XML <Age>25</Age> A up</rating> to <movie> Fugitive, The <Salary>20000</Salary> B Fugitive, The</movie> </person> <person> <Name>B</name> <Age>62</Age> <Salary>140000</Salary> </person>

7 Publishing Relational Databases Name A B Age Salary <person> <Name>A</name> <Age>25</Age> </person> Users/Applications see a uniform XML view Exchange data with other applications Querying XML is easier? Problem: What is a good XML schema for a relational schema?

8 XML for Data Modeling Location GPS) GPS gps Location Bstation*) Bstation Location (GPS Bstation*))

9 XML as a logical data model Location location (GPS Bstation*)) Use data modeling features provided by XML Union types Recursive types Ordered relationships Easier to Query? Problems: What is a good XML schema for an application? How do we store the data in relational databases?

10 XML for data integration Source1 Browse Query, Update XML Wrapper Source2 SourceN

11 Database Design Stages Application Requirements Conceptual Design Conceptual Schema Logical Design Logical Schema Physical Design Physical Schema

12 Logical Data Model and Redundancy Bad Design Student_Professor Professor sname BS advisor age pname age SD CS MXM 40 MXM 40 YC EE MXM 40 Student sname BS advisor SD CS MXM YC EE MXM Good Design Person name address city state zip AV A1 Los An CA AN A2 Los An CA 90034

13 What is a Data Model? Structural Specification Constraint Specification Operations

14 Entity Relationship (ER) Model Structures: Entity Types, Relationship Types Constraints: Cardinality constraints Professor (0,*) (1,1) Advisor Student pname age year sname

15 Relational Model Structures: Relations Constraints: Key, Foreign Key Professor pname age Student sname advisor Key Constraints: Key (Professor) = <pname> Key (Student) = <sname> Foreign Key Constraints: Student (advisor) references Professor (pname)

16 Specifying Structures for XML G = (N, T, P, S) N = {Book, Author, Publisher, #PCDATA} T = {book, author, publisher, pcdata} S = {Book} Book book (Author +, Publisher) Author author (#PCDATA) Publisher publisher (@name::string) #PCDATA pcdata (ε) Regular Tree Grammar Every production rule is of the form A a X A N, a T, X is a regular expression over N book author author publisher J. E. Hopcroft J. D. Ullman name= "Addison- Wesley"

17 XML Schema Language Proposals W3C DTD: local tree grammar W3C XML Schema: single type tree grammar ISO/OASIS RELAX NG: full-fledged regular tree grammar

18 Properties of different Regular Tree Grammar classes Expressiveness Regular tree grammar strictly more expressive than single type tree grammar Single type tree grammar strictly more expressive than local tree grammar Closure properties Regular tree grammar closed under union, intersection and difference Single type tree grammar/local tree grammar closed only under intersection Type assignment Type assignment can be ambiguous for regular tree grammar. Type assignment is unambiguous for local tree grammar/single type tree grammar.

19 Ambiguous Type Assignment G = (N, T, P, S) N = {Book, Author1, Author2, Publisher, #PCDATA} T = {book, author, publisher, pcdata} S = {Book} Book book (Author1*, Author2*, Publisher) Author1 author (#PCDATA) Author2 author (#PCDATA) Publisher publisher (@name::string) #PCDATA pcdata (ε) book author author publisher J. E. Hopcroft J. D. Ullman name= "Addison- Wesley"

20 Constraint Specification for XML why? If we represent all relationships only by hierarchies, then the logical model will have redundancy. Person (0,*) (0,*) Review Book name zip rating ISBN title year What constraint specification? Key, Foreign Key ID/IDREF

21 Specifying Constraints for XML: Example library name= "RRM" person person name= "CZ" book book paper ISBN= "I1" title= "T1" BID= "B1" ISBN= "I2" title= "T2" BID= "B2" title= "T1" journal ="J1" PID= "P1" review review review review article ="P1" rating ="9" article ="B2" rating ="9" article ="B1" rating ="9" article ="B2" rating ="10"

22 Specifying Constraints for XML Keys are specified using (rel, sel, field) rel is relative axis sel is selector axis field is a set of path expressions For any element that belongs to rel, sel will give a set of elements. For this set of elements, field is the key. rel and sel can be types or path expressions Foreign keys are specified as (rel 1, sel 1, field 1 ) references (rel 2, sel 2, field 2 )

23 Constraint Specification Proposals W3C XML Schema Relative axis = type Selector axis = path expression Keys for XML WWW10 Relative axis = path expression Selector axis = path expression UCM WWW10 No relative axis Selector axis = type

24 Our proposal Relative axis = type Selector axis = type IDREF and IDREFS identify target types

25 Example library name= "RRM" person person name= "CZ" book book paper ISBN= "I1" title= "T1" BID= "B1" ISBN= "I2" title= "T2" BID= "B2" title= "T1" journal ="J1" PID= "P1" review review review review article ="P1" rating ="9" article ="B2" rating ="9" article ="B1" rating ="9" article ="B2" rating ="10" (Library, Person, (Library, Book, (Library, Paper, (Person, references (Book Paper)

26 Why use XML as logical data model?

27 Path Expressions vs Joins Professor pname age MXM CZ university Student sname year advisor pname="mxm" age="40" professor professor pname="cz age="55" SD 1998 MXM student student student YC FC MXM CZ sname ="SD" year= "1998" sname ="YC" year= "2000" sname ="FC" year= "1999" Student (advisor) references Professor (pname) Query: Give names of students of professors of age 40 π name ((σ age=40 (Professor)) Student) professor

28 Union Types - attributes name AV AN Person city state Los A CA null null zip null Person person university person person name ="AV" city= "Los A" state = CA name ="AN" zip= 90034

29 Union Types - Relationships name RRM CZ Person zip Person person (Book* Paper*)) ISBN I1 I2 Book title T1 T2 author RRM RRM library title T1 Paper journal J1 author CZ name= "RRM" person zip= "01609" name= "CZ" person zip= "90095" book book paper ISBN= "I1" title= "T1" ISBN= "I2" title= "T2" title= "T1" journal ="J1"

30 Union Types - Relationships Conference name ER Journal name TOIT venue Chicago publisher ACM title T1 T2 inst WPI UCLA Paper conf ER null journal null TOIT Conference conference Paper*) Journal journal Paper*) library name= "ER" conference venue= "Chicago" name= "TOIT" journal publisher ="ACM" paper paper title= "T1" inst= "WPI" title= "T2" inst= "UCLA"

31 Recursive Types Assembly assembly name seat superpart frame qty 1 name=bike qty=0 part tire frame wheel bike 1 1 name=frame qty=1 part part name=wheel qty=2 wheel bike bike null 2 0 name=seat qty=1 part part name=tire qty=1 WITH RECURSIVE SubPart (name) AS (SELECT name FROM Assembly WHERE superpart=bike) UNION (SELECT R2.name FROM SubPart R1, Assembly R2 WHERE R2.superPart = R1.name) SELECT * FROM SubPart Query: What are subparts of bike? part[@name=bike]//part/@name

32 IDREF vs Foreign Keys Professor name age MXM 40 CZ 55 Student name year advisor name= "MXM" professor age= "40" PID= "P1" university name= "CZ" professor age= "55" PID= "P2" SD 1998 MXM student student student YC FC MXM CZ name= year= "SD" "1998" PRef ="P1" name= year= "YC" "2000" PRef ="P1" name= "FC" year= "1999" PRef ="P2" Student (advisor) references Professor references (Professor) Query: Give names of students of professors of age 40

33 IDREF as union of foreign keys ISBN I1 I2 title T1 Book title T1 T2 journal J1 name RRM Paper CZ Review Person zip name book paper rating RRM null T1 9 CZ I1 null 9 CZ I2 null 10

34 IDREF as union of foreign keys library name= "RRM" person zip= "01609" name= "CZ" person zip= "90095" book book paper ISBN= "I1" title= "T1" BID= "B1" ISBN= "I2" title= "T2" BID= "B2" title= "T1" journal ="J1" PID= "P1" review review review article ="P1" rating ="9" article ="B1" rating ="9" article ="B2" rating references (Book Paper)

35 Conceptual Model: ERex (ER extended for XML)

36 Entity Relationship (ER) model Entity Types, Relationship Types and their attributes Professor (0,*) (1,1) Advisor Student pname age year sname

37 Object Role Modeling (ORM) closer to natural language sentences attributes/relationships are expressed uniformly using roles Student (sname) age has Professor (pname) is advisor of for Years (year)

38 Unified Modeling Language (UML) Modeling software systems Class Diagrams, Association Classes Professor pname age * Student sname Advisor years

39 From ER model Binary 1:n relationships Professor (0,*) (1,1) Advisor Student pname age year sname Person (0,*) (0,*) Review Book Binary m:n relationships name zip rating ISBN title year

40 From ER Model quantity N-ary relationships Company (1,*) (0,*) Supplies City (0,*) name Product name name superpart (0,*) Contains Recursive relationships Part subpart (0,1) name qty

41 Ordered Relationships Person (0,*) (1,1) Author Book name zip ISBN title Person name zip Ullman RRM Book ISBN title order B1 DB 2 B2 Aut 1 B3 MMSL 1 author Ullman Ullman RRM

42 Categories and Set Constraints name Person PersonCity PersonZip PersonCity PersonZip = PersonCity PersonZip = Person city state zip

43 Categories name Person (0,*) Review (0,*) Article Book Paper

44 Set constraints on Roles (0,*) personbook AuthorOf (1,1) Book name Person (0,*) personpaper AuthorOf (1,1) Paper personbook personpaper =

45 Set Constraints on Roles name venue Conference (0,*) Presented (0,1) confpaper Paper title name publisher Journal (0,*) Published (0,1) journalpaper confpaper journalpaper = confpaper journalpaper = Paper

46 Translating ERex schemas to XML schemas

47 System Architecture ERex Schemas Translator XML Schemas Final XML Schema Schema Designer

48 1:n relationships Professor (0,*) (1,1) Advisor Student pname age year sname

49 Representing 1:n Relationships - subelement Professor name RRM CZ name MM YC FC age Student year advisor RRM RRM CZ Student (advisor) references Professor (name) name="rrm" age="62" name= "MM" student year= "1998" professor name= "YC" university student year= "2000" professor name= "FC" student University university (Professor*) Professor professor Student*) Student student (University, Professor, <@name>) (University, Student, <@name>) year= "1999" name="cz" age="55"

50 Representing 1:n Relationships - IDREF Professor name age university RRM 62 professor professor CZ 55 Student name= "RRM" age= "62" PID= "P1" name= "CZ" age= "55" PID= "P2" name year advisor student student student MM YC RRM RRM name= year= "MM" "1998" PRef ="P1" name= year= "YC" "2000" PRef ="P1" name= year= PRef "FC" "1999" ="P2" FC 1999 CZ Student (advisor) references Professor (name) University university (Professor*, Student*) Professor Student (University, Professor, <@name>) (University, Student, references (Professor)

51 Representing 1:n Relationships foreign keys name RRM CZ name MM YC FC Professor age Student year advisor RRM RRM CZ Student (advisor) references Professor (name) student name= year= "MM" "1998" name= "RRM" professor advisor ="RRM" age= "62" university student name= year= "YC" "2000" name= "CZ" advisor ="RRM" professor age= "55" name= "FC" student year= "1999" advisor ="CZ" University university (Professor*, Student*) Professor (University, Professor, (University, Student, (University, Student, references (University, Professor,

52 m:n relationships Person (0,*) (0,*) Review Book name rating ISBN title

53 Representing m:n relationships Person name RRM CZ pname RRM RRM CZ CZ Book ISBN I1 I2 I3 Review ISBN I3 I2 I1 I1 9 9 title T1 T2 T3 rating 9 10 name= "RRM" ISBN= "I1" article ="B3" book title= "T1" review BID= "B1" rating ="9" person ISBN= "I2" article ="B2" book title= "T2" review rating ="9" BID= "B2" library ISBN= "I3" book title= "T3" BID= "B3" article ="B1" Library library (Person*, Book*) Person person (@name, Review*) Book Review review review rating ="9" person article ="B2" review name= "CZ" rating ="10" Review (pname) references Person (name) Review (ISBN) references Book (ISBN) (Library, Person, <@name>) (Library, Book, <@ISBN>) (Person, Review, references (Book)

54 Representing m:n relationships Person name RRM CZ pname RRM RRM CZ CZ Review ISBN I3 I2 I1 I1 Book ISBN I1 I2 I3 rating title T1 T2 T3 name= "RRM" ISBN= "I1" article ="B3" book title= "T1" review BID= "B1" rating ="9" person book review library book Query: What is the rating given by RRM for book T1 π rating ((σ title=t1 (Book)) (σ pname=rrm (Review))) review person person[@name=rrm]/review[@article Book/title=T1]/@rating ISBN= "I2" article ="B2" title= "T2" rating ="9" BID= "B2" ISBN= "I3" title= "T3" BID= "B3" article ="B1" rating ="9" article ="B2" review name= "CZ" rating ="10"

55 N-ary relationships quantity Company (1,*) (0,*) Supplies City (0,*) name Product name Root root (Company*, Product*, City*) Company company Supply + ) Supply Product product City city name (Root, Company, <@name>) (Root, Product, <@name>) (Root, City, references references (City)

56 Recursive Relationships superpart (0,*) Contains Part subpart (0,1) assembly Assembly name qty name=bike qty=0 part name seat tire superpart frame wheel qty 1 1 name=frame qty=1 name=seat qty=1 part part part part name=wheel qty=2 name=tire qty=1 frame wheel bike bike bike null Assembly assembly (Part*) Part part Part*) (assembly, part, <@name>)

57 Ordered Relationships Person name Ullman zip library RRM name="ullman" zip="95123" person person name="rrm" zip="90095" Book book book book ISBN B1 title DB order 2 author Ullman ISBN= "B1" title= "Aut" ISBN= "B2" title= "DB" ISBN= "B3" title= "MMS" B2 Aut 1 Ullman B3 MMS 1 RRM Library library (Person*) Person person Book*) Book book (Library, Person, <@name>) (Library, Book, <@ISBN>)

58 Categories and set constraints name Person PersonCity PersonZip PersonCity PersonZip = PersonCity PersonZip = Person city state zip Person person

59 Categories and Set Constraints name Person Review (0,*) (0,*) Root root (Person*, Book*, Paper*) Person person Review*) Review (Root, Person, (Person, references (Book Paper) Article Book Paper

60 Set constraints on Roles (0,*) personbook AuthorOf (1,1) Book name Person (0,*) personpaper AuthorOf (1,1) Paper personbook personpaper = Person person (@name, (Book* Paper*))

61 Set Constraints on Roles name venue Conference (0,*) Presented (0,1) confpaper Paper title name publisher Journal (0,*) Published (0,1) journalpaper confpaper journalpaper = confpaper journalpaper = Paper Conference conference Paper*) Journal journal Paper*)

62 Converting ERex XML Goals Maximize relationships represented using subelement. Others try to represent using IDREF

63 Algorithm: ERex XML A non-terminal symbol for each entity type with key m:n relationship n-ary relationship Root non-terminal symbol Represent attributes Represent relationships and identify top nodes 1:1 and 1:n relationships m:n relationships n-ary relationships Identify key and IDREF constraints.

64 Review (0,*) (0,*) rating Article (0,*) personbook Author Of (1,1) Book Paper name Person (0,*) personpaper Author Of (1,1) ISBN btitle year ptitle year journal PersonCity PersonZip personbook personpaper = personbook personpaper = Person city state zip PersonCity PersonZip = PersonCity PersonZip = Person

65 Review (0,*) (0,*) rating Article (0,*) personbook Author Of (1,1) Book Paper name Person (0,*) personpaper Author Of (1,1) ISBN btitle year ptitle year journal PersonCity PersonZip city state zip N = {Root, Person, Book, Paper, Review} Book Paper Person person Person person (Book* Paper*) Person person (Review*) Review review Root root (Person*) (Root, Person, <@name>) (Root, Book, <@ISBN>) (Root, Paper, <@ptitle>) (Person, Review, references (Book Paper)

66 Conclusions ERex Schemas Translator XML Schemas Final XML Schema Schema Designer Obtained good XML Schema from ERex schemas

67 Part II: Translation between Relational and XML models.

68 Why publish relational databases as XML? Provide an XML view for our legacy data Users/Applications can query our data over the web using standards. Easier to query XML than legacy (relational) data? Convert our legacy data to XML We can exchange data with applications. Store data in XML databases? Easier to query?

69 Application Scenarios Schema Matching Problem Given a relational schema and an XML schema by a standards body, how do we map this relational schema to XML? Tools such as XML Extender from IBM, Clio (University of Toronto and IBM), MS SQL Server Schema Mapping Problem Given a relational schema, how do we come up with a good XML schema?

70 Schema Matching: MS SQL Server Architecture COPYRIGHT NOTICE. Copyright 2003 Microsoft Corporation, One Microsoft Way, Redmond, Washington U.S.A. All rights reserved.

71 Schema Mapping Problem Name A B Age Salary <person> <Name>A</name> <Age>25</Age> </person> Users/Applications see a uniform XML view Exchange data with other applications Store XML in native XML databases Querying XML is easier? Problem: What is a good XML schema for a relational schema?

72 Goals XML Schema should maintain constraints. Resulting XML should not introduce redundancies. Most relationships can be navigated using path expressions, rather than joins. Minimal user interaction: Our translator should suggest good XML schemas to the database designer.

73 System Architecture RDB NeT CoT XML Schemas Final XML Schema Schema Designer

74 Related Work XML-DBMS Template driven mapping language SilkRoute Declarative Query Language (RXL) for viewing relational data as XML Xperanto User specifies query in XML Query Language

75 Algorithms Naïve FT (Flat Translation) Consider relational data NeT (Nesting-based Translation) Consider relational schema CoT (Constraint-based Translation)

76 FT: Flat Translation 1 : 1 mapping from relational to XML Idea A type (non-terminal) corresponding to every relation Attributes of a relation form attributes of the type Keys and foreign keys are preserved

77 FT: Flat Translation - Example name RRM CZ name MM YC FC Professor age Student year advisor RRM RRM CZ Student (advisor) references Professor (name) student name= year= "MM" "1998" name= "RRM" professor advisor ="RRM" age= "62" university student name= year= "YC" "2000" name= "CZ" advisor ="RRM" professor age= "55" name= "FC" student year= "1999" University university (Professor*, Student*) Professor professor Student (University, Professor, <@name>) (University, Student, <@name>) (University, Student, <@advisor>) references (University, Professor, <@name>) advisor ="CZ"

78 NeT: Nesting-based Translation Idea: Make use of non-flat features provided by XML: represent repeating groups using *, +

79 NeT: Example Course (cname, prof, text) cname Algorithms Algorithms Algorithms Algorithms prof Gafni Gafni Majid Majid text Udi Manber CLR Udi Manber CLR

80 NeT: Example Course (cname, prof +, text) cname Algorithms Algorithms prof {Gafni, Majid} {Gafni, Majid} text Udi Manber CLR Course (cname, prof +, text + ) Cname Algorithms prof {Gafni, Majid} text {Udi Manber, CLR}

81 NeT: Example university course course course course University university (Course*) cname="algorithms" prof="gafni" text="udi Manber" cname="algorithms" prof="gafni" text="clr" cname="algorithms" prof="majid" text="udi Manber" cname="algorithms" prof="majid" text="clr" university ValueRatio=12/5 University university (Course*) Course course Prof+, Text+) Prof prof (#PCDATA) Text text (#PCDATA) #PCDATA pcdata (ε) cname="algorithms" prof prof course text text (University, Course, prof, text>) Gafni Majid Udi Manber CLR

82 NeT: Example Person (name, city, state, zip) name city state zip MM Los Angeles CA AN Los Angeles CA Person (name +, city, state, zip) {MM, AN} Los Angeles CA 90034

83 NeT: Example directory directory person person city="los Angeles" state="ca" person zip="90034" name="mm" city="los Angeles" state="ca" zip="90034" name="an" city="los Angeles" state="ca" zip="90034" Directory directory (Person*) @zip) (directory, person, ValueRatio=8/5 name MM name AN Directory directory (Person*) Name*) Name name (#PCDATA) (directory, person,

84 NeT: Summary Consider Table t with column set C. Nesting on column X is defined as: Any two tuples with the same values for (C X) will be combined to one tuple Observation: We need to nest only on key columns Advantages of NeT NeT removes redundancy if relation is not in 4NF NeT provides more intuitive XML schemas with less redundancy

85 NeT: Experimentation Test Set #attr #tuples ValueRatio #nested attributes Time (sec) Balloons Hayes Bupa Balance TA_Eval Car Flare

86 CoT: Constraint-based Translation Translating relational schema Idea: Use foreign key constraints and our knowledge of how to represent relationships to come up with better XML models.

87 CoT: Step 1 university Professor pname RRM age 62 pname="rrm" age="62" professor Student student student student sname MM YC YC advisor RRM RRM RRM cname DBs DBs QSs Student (advisor) references Professor (pname) ValueRatio=11/8 sname= "MM" cname= "DBs" University university (Professor*) Professor Student*) Student student (University, Professor, <@pname>) (Professor, Student, sname= "YC" cname= "DBs" sname= "YC" cname= "QSs"

88 CoT: Step 2 Professor pname RRM sname MM YC YC age 62 Student advisor RRM RRM RRM Course cname DBs QSs cname DBs DBs QSs since pname="rrm" age="62" University university (Professor*, Course*) Professor professor Student*) Student student Course (University, Professor, <@pname>) (University, Course, <@cname>) (Professor, Student, CRef::IDREF references (Course) professor university course course Student (advisor) references Professor (pname) Student (cname) references Course (cname) sname= "MM" student CRef= "c1" student sname= "YC" CRef= "c1" student sname= "YC" cname= "DBs" since= "1979" CRef= "c2" CID= "c1" cname= "QSs" since= "1962" CID= "c2" ValueRatio=15/14

89 CoT: Example Student (SID, name, advisor) Emp (EID, name, projname) Prof (EID, name, teach) Course (CID, title, room) Dept (dno, mgr) Proj (pname, pmgr) student student proj Student (advisor) references Prof (EID) Emp (projname) references Proj (pname) Prof (teach) references Course (CID) Prof (EID, name) references Emp (EID, name) Dept (mgr) references Emp (EID) Proj (pmgr) references Emp (EID) course prof emp dept

90 student student proj course course prof Top nodes emp dept (University, Course, (University, Emp, (University, Prof, (University, Student, (University, references (Emp) University university (Course*, references Prof*) (Proj) Prof prof Student*) Student @EmpID, Dept*, Proj*) Proj Dept dept

91 CoT Experimentation Ran on TPC-H data Value ratio > 100/88 (size decreased by more than 12%)

92 Conclusions RDB NeT CoT XML Schemas Final XML Schema Schema Designer We obtained good XML schemas from relational schemas Constraints are maintained Redundancies are decreased Most relationships can be navigated using path expressions. Minimum user interaction

93 Storing XML data in relational databases

94 Options for storing XML data Store in relational databases Relational databases are robust and efficient (IBM XML Extender, Oracle, MS SQL Server) Store in native XML databases More efficient for XML than relational databases (Natix, exist, Tamino) Store in a combination of both Structured portion of XML data in relational database, and unstructured portion in native XML store.

95 Related Work STORED No Schema Use data mining techniques to find structured and frequent patterns These are stored in relational DB, others in semistructured overflow store Drawback: Requires integration of relational DB and semi-structured store Storing paths One relation for storing nodes, one for storing edges. Drawback: Type information is lost.

96 Type-based relational storage Jayavel Shanmughasundaram Several key ideas such as schema simplification, inlining, handling recursion LegoDB Use the query workload to come up with an efficient relational schema.

97 Main features The entire XML document is shredded and stored in a relational database. All semantic constraints in XML schema are not captured in relational schema. We do not discuss how operations on XML are translated to SQL.

98 Why not capture all constraints? A a (@b, ((@c, D)* (@e, F)*)) (Root, A, <@b>) D (@aref) F (@aref) Semantic constraints lost if D refers to an A, then the should be non-null if F refers to an A, then the should be non-null

99 NF2 representation of regular tree grammars Eliminate

100 Schema Simplification A a (@d, (B, C, B)*) A a (@d [1, 1], (B [2, 2], C [1, 1]) [0, *]) A a (@d [1, 1], B [2, 2] [0, *], C [1, 1] [0, *]) A a (@d [1, 1], B [0, *], C [0, *]) A a (@d, B*, C*) Semantic information lost The number of B s is two times the number of C s for every C, there is a B that occurs before it, and one that occurs after it.

101 Inlining Conf Venue) Venue @country) Why inlining? Lesser joins, hence more efficient

102 Mapping Collection Types Conf Paper*) Paper @confref) Separate relation with foreign key for every collection type

103 IDREF attribute Person Review*) references (Book Paper) Book Paper

104 Recursion using? A a (@d, A?) A (@d, ARef) ARef refers to the child of A, and can be null A a (@d, ARef) ARef refers to the parent of A and can be null.

105 Recursion using * A a (@d, A*) A (@d, ARef) ARef refers to the parent of A, and can be null

106 Recursion General technique For every cycle, we must have a separate relation Algorithm For every strongly connected component, define a separate relation for one of the types. In a strongly connected component, if there is a type which can be children of multiple types, then define a separate relation for that type.

107 Capturing Order in the Document Through order attributes Corresponding to each type in XML schema, say A, we have an associated order attribute, say aorder Conf Venue) Venue @venueorder)

108 Conclusions XML schema with no recursion can be translated to relational schema with no nulls. XML schema with recursion cannot be translated to relational schema with no nulls. If recursion, separate relation needed for every cycle. All semantic constraints in XML cannot be captured in relational schema. XML resulting from CoT can be translated to the original relational schema; all semantic constraints are maintained.

109 Open Problems Standards specification: What structural and constraint specification schemes for XML are needed for database applications? XML used for text/document publishing: Keyword Search in XML documents Storing data consisting of structured and unstructured portions: integrating relational and XML stores.

110 Open Problem (contd ) Translating operations in XML model to underlying sources (relational) Use annotated schema (MS SQL Server) Use implicit annotations (LegoDB) Query minimization: When we do automatic translation, we might perform unnecessary joins?