Relational Databases for Querying XML Documents: Limitations and Opportunities

1
Relational Databases for Querying XML
DocumentsLimitations and Opportunities

• Presented by
• Yi Lu

2
Introduction

• XML is fast emerging as the dominant standard for
  representing data in the World Wide Web.
• The initial purpose of XML is to enhance the
  ability of exchanging data over the Internet.
• It raises a problem how to query the contents of
  the XML documents.

3
Approaches for querying XML documents

• Use semi-structured query languages and query
  evaluation techniques.
• Use relational database to store and query XML
  documents.
• Native XML repositories, e.g., Software AGs
  Tamino, eXcelons XIS. (summarized by Lus paper)

4
Processes that used in the relational approach

• Process a XML DTD to generate a relational schema
• Parse XML documents conforming to DTDs and load
  them into tuples of relational tables in a
  standard commercial DBMS (DB2, Oracle)
• Translate semi-structured queries over XML
  documents into SQL queries over relational
  database
• Convert the results back to XML format.

5
Outline of the talk

• XML background
• Mapping XML DTD to relational schema
• Basic inlining techniques
• Shared inlining techniques
• Hybrid inlining techniques
• Some experiments
• Translating semi-structured queries into SQL
  queries
• Converting results to XML format.
• Conclusions and future work

6
XML background

• Extensible Markup Language
• DTDs and other XML schemas
• DCD (Document Content Descriptor)
• XML Schemas
• Semi-structure query languages
• XML-QL, Lorel, UnQL, XQL
• notion of path expressions for navigating nested
  structure of XML

7
XML Query Languages

• XML-QL Use nested XML-like structure

8
XML Query Languages

• Lorel more like SQL
• In this paper, the combination of XML-QL Lorel
  is used to do demonstration

9
Outline of the talk

• XML background
• Mapping XML DTD to relational schema
• Basic inlining techniques
• Shared inlining techniques
• Hybrid inlining techniques
• Some experiments
• Translating semi-structured queries into SQL
  queries
• Converting results to XML format.
• Conclusions and future work

10
Mapping DTDs to relational schemas

• Simplifying DTDs
• Creating and inlining DTD graphs
• Generating relational schemas

11
Simplifying DTDS

• Why?
• DTDs can be complex
• Generating relational schemas would be unwieldy
• Can simplify DTD and still generate relational
  schema that can store and query documents
  conforming to the original DTD.
• Transformation preserves semantics of
• one or many
• null or not null
• Loses some information about relative orders of
  the element
• can be captured when a specific XML doc is loaded
  into relational schema?

12
Simplifying DTDS (Cont)

• Flatting transformations
• Simplification transformations

13
Simplifying DTDS (Cont)

• Grouping transformations
• All operators are transformed to
• A DTD simplification example
• lt!ELEMENT a((bce)?,(e?f?,(b,b))))gt
• Transformed to lt!ELEMENT a(b,c?,e,f)gt

14
Motivation for special schema conversion

• Relational schema
• derived from data model such as ER-model
• clear separation between Entity and Attribute
• Try mapping DTDs element and attribute to ERs
  entity and attribute
• no correspondence
• lead to excessive fragmentation of the document

15
DTD graph

• Represents the structure of DTD
• Nodes
• Elements appear exactly once
• Attributes appear as many times as they appear
• Operators appear as many times as they appear
• Cycles in the DTD graph indicate the presence of
  recursion

16
An example used in the paper
17
DTD graph
18
The Basic inlining technique

• Creating a relation with DTD graph
• All elements descendents are inlined into that
  relation
• Exception
• children directly below node are made into
  separate relations
• each node having a backpointer edge pointing to
  it is made into a separate relations

19
The Basic inlining technique

• Attributes are named by the path from the root
• Each relation has an ID field
• key of the relation
• All relation corresponding to element nodes
  having a parent have a parentID field
• foreign key

20
The Basic inlining technique
21
The Shared inlining technique

• Attempts to avoid drawbacks of Basic
• Principal idea
• identify element nodes that are represented in
  multiple relations
• share them by creating separate relations

22
The Shared inlining technique

• Creating a relation that all elements in the DTD
  graph whose nodes have in-degree greater than
  one
• in-degree of 1 inlined
• in-degree of 0 separate relation is created
• Elements below node are made into separate
  relations

23
The Shared inlining technique
24
The Hybrid inlining technique

• Inlines some elements that are not inlined in
  Shared
• Inlines elements with in-degree greater than one
• that are not recursive
• reached through a node

25
The Hybrid inlining technique
26
A qualitative evaluation of the Basic, Shared and
Hybrid techniques

• Evaluation Metric
• Major concern efficiency of query processing
• Average number of SQL joins required to process
  path expressions of a certain length N
• Measurements
• The average number of SQL queries generated for
  path expressions of length N
• The average number of joins in each SQL query for
  path expressions of length N
• The total average number of joins in order to
  process path expressions of length N
• Concentrate on comparisons between Shared and
  Hybrid

27
Evaluation results

• Hybrid eliminates large number of joins for some
  DTDs
• Hybrid requires more SQL queries than using
  Shared for some DTDs
• Shared always produces at least number of join
  per SQL query as Hybrid
• Hybrid always produces at least the number of SQL
  queries as Shared

28
Outline of the talk

• XML background
• Mapping XML DTD to relational schema
• Basic inlining techniques
• Shared inlining techniques
• Hybrid inlining techniques
• Some experiments
• Translating semi-structured queries into SQL
  queries
• Converting results to XML format.
• Conclusions and future work

29
Translating semi-structured queries to SQL queries

• Semi-structured QL have more flexibility than
  SQL, allow path expression with various operators
  and wild cards
• Converting queries with simple path expression to
  SQL
• Converting simple recursive path expressions to
  SQL
• Converting arbitrary path expression to simple
  recursive path expressions

30
Converting queries with simple path expression to
SQL

• Relation corresponding to start of the root path
  expression are identified
• Add it to the from clause of the SQL query
• Path expressions are translated to joins among
  relations

31
Converting simple recursive path expressions to
SQL

• Determine the initialization of the recursion and
  the actual recursive path expression
• Ask for the names of all editors reachable
  directly or indirectly from the monograph with
  title Subclass Cirripedia

32
Converting arbitrary path expression to simple
recursive path expressions

• Path expression can be of arbitrary complexity
• ask for all the name elements reachable directly
  or indirectly through monograph
• Method
• take path expression appearing in such query
• translate them into possibly many simple path
  expressions

33
Outline of the talk

• XML background
• Mapping XML DTD to relational schema
• Basic inlining techniques
• Shared inlining techniques
• Hybrid inlining techniques
• Some experiments
• Translating semi-structured queries into SQL
  queries
• Converting results to XML format.
• Conclusions and future work

34
Converting relational results to XML

• Explorer How results by SQL queries can be
  converted XML documents
• It is difficult to constructing arbitrary XML
  result, and it is the main drawback in using
  current relational approach
• use XML-QL as the illustrative query language, it
  provides XML structuring constructs

35
Simple structuring
The first and last name of all the authors of
books.
36
Tag variables

• Generating a relational query that contains tag
  value as an element of the result tuple.
• Ask for The names of authors of all
  publications, nested under a tag specifying the
  type of publication

37
Grouping

• Ask all the publications of an author to be
  grouped together, and within this structure,
  requires the titles of publications to be grouped
  by the type of publication.
• Two approach can be used
• The relational database order the result tuples
  by last name and then by publication type, and
  scan the result to construct the XML document. It
  is showed in the figure.
• get an unordered set of tuples and do a grouping
  operation outside of database engine, by last
  name and by type.

38
Grouping

• Treating tag variables as attribute in the result
  relation provided a way of uniformly treating
  the contents of the result XML document
• Some relational database functionality is either
  not fully exploited or is duplicated outside.

39
Converting other element type

• Complex element construction
• Mainly concern about the set values
• Heterogeneous Results
• The different queries can be handled in different
  ways, and then the results can be merged together
• Nested Queries
• Using outer join to construct the association
  between a query and a sub-query

40
Conclusions

• Study the virtues and limitation of the
  relational model for processing queries over XML
• The advantages is reusing relational database
  technology which has high performance. The paper
  shows the possibility to handle most queries on
  XML document using a relational DB
• Limitations
• Awkward complex XML constructs in their results
• Inefficient Fragmentation causes too many joins
  in the evaluation of simple queries

41
Future work

• Support for sets
• Untype/variable-type references
• Information retrieval style indices
• Flexible comparisons operators
• Multiple-query optimization execution
• More powerful recursion

42
Questions?