TIMBER: A Native XML Database

1
TIMBER A Native XML Database

• Author H.V. Jagadish, etc.
• Presenter Min Lu
• Date Apr 5, 2005

2
Introduction

• Growing XML XML repository
• New Approach Native XML DB
• TIMBER Tree-structured native XML database
  Implemented at the University of Michigan by
  Bright Energetic Researchers

3
Topics of Discussion

• Motivation
• TIMBER Architecture
• Tree Algebra (TAX)
• Query Optimization
• Conclusion

4
Motivation
5
Motivation

• XML Characteristics
• Tree structured - elements can be structurally
  
• related and these relationships are
  meaningful
• Flexibility
• Map XML to Relational DB
• Unnormalized relational representation
• Or a large number of tables

6
Motivation

• Native XML DB
• Tamino - a commercial one
• Natix - a native XML data management system,
  designed for storing and processing XML data.
• Timber on Shore storage manager.

7
Topics of Discussion

• Motivation
• TIMBER Architecture
• Tree Algebra (TAX)
• Query Optimization
• Conclusion

8
TIMBER Architecture

• Shore
• Disk memory
• management
• Buffering
• Concurrency
• control

(Shore)
9
TIMBER Architecture Data Flow
Interface
Interface
Interface
Parse tree
One node at a time
Internal representation
(Shore)
10
TIMBER Architecture Query Flow
Operator tree
Call
Call
Call
Call
(Shore)
11
Nodes in TIMBER

• One node for each element
• All attributes clubbed into one node
• Content of element pulled into a child node
• Processing instruction, comments are simply
  ignored

12
Node Labels

• The determination of PC, AD relationships is a
  frequent operation
• Label each node with a triple
• Start, end, level (S, E, L)

13
Triple Labels for AD PC

• AD (S1, E1, L1) - (S2, E2, L2)
• ltgt S1ltS2 E1gtE2
• ex. (1.0, 9.0, 1) (3.0, 6.0, 5)
• PC (S1, E1, L1) - (S2, E2, L2)
• ltgt S1ltS2 E1gtE2 L1L2-1
• ex. (1.0, 9.0, 1) (2.0, 8.0, 2)

14
Triple Label Benefits

• Updates no re-labeling
• Use Double value to leave gaps for new nodes
• Serves as a node identifier
• Store nodes by the start labels to cluster their
  sub-elements together with them

15
Topics of Discussion

• Motivation
• TIMBER Architecture
• Tree Algebra (TAX)
• Query Optimization
• Conclusion

16
Tree Algebra (TAX)

• Set-at-a-time for efficiency
• Bulk algebra input one or more sets of trees and
  output a set of trees
• Pattern tree the portion of interest
• Witness tree bears witness to the success of the
  pattern match on the input tree

17
Pattern Tree Witness Tree
A
C
B
18
Operators in TAX

• Algebra Operations developed
• Selection, Projection, Product,
• Set union, Set difference,
• Renaming, Reordering, Grouping
• The core of XQuery can be parsed to TAX operators

19
Projection Operator in TAX

• Input C collection of trees
• Parameter P pattern tree
• Parameter PL projection list
• (the info to keep in the output)

20
Topics of Discussion

• Motivation
• TIMBER Architecture
• Tree Algebra (TAX)
• Query Optimization
• Conclusion

21
Query Optimization

• Consider the join between faculty node and
  secretary node first, then join the result with
  RA node.
• Join faculty node with RA node first, then, join
  the result with secretary node.

22
Query Optimizer

• Query optimizer enumerates all evaluation plans,
  estimate their costs, then choose the optimal
  one.
• An algorithm FP_Optimization for finding the best
  evaluation plan.

23
Case Study for Query Optimization

• Consider the query against the DB mBench 0.1x
  data set with about 130,000 nodes

A
B
A
D
B
F
D
F
C
E
C
E
G
G
24
Query Optimization
25
Performance Study
26
Topics of Discussion

• Motivation
• TIMBER Architecture
• Tree Algebra (TAX)
• Query Optimization
• Conclusion

27
Conclusion

• A comprehensive set-at-a-time query processing
  ability in a native XML store, with all the
  standard components of relational query
  processing
• New access methods have been developed to
  evaluate queries from XML
• New cost estimation and query optimization
  techniques have been developed.

28
Work to be Done

• Currently all processing instructions, comments,
  and such are simply ignored.
• - An extra child node of the element node with
  all such
• data needs to be created.
• TIMBER was developed when XQuery didnt support
  updates.
• - 11th Feb 2005 First Public Working Draft of
  the XQuery
• Update Facility Requirements
• - A parser has to be implemented to support
  updates.
• During an extremely localized sequence of
  inserts, the Start End labels become an issue.

29
Questions?

• Thank you!