XML Storage

1
XML Storage

• Maya Ramanath
• Database Systems Lab
• SERC, IISc

2
How should we store XML?

• File system
• Text files - differently represented during
  processing
• Text files - with additional indexes
• Database systems
• Relational
• Object-oriented, etc.

3
File System

• Advantages
• Easy to store
• Document granularity is maintained
• Disadvantages
• Query processing

4
Database Systems

• Advantages
• Well understood technology
• Optimized for query processing
• Disadvantages
• XML does not directly map to any data model
• Not amenable to a rigid schema

5
Example

• ltSTUDENTgt
• ltNAMEgt Maya lt/NAMEgt
• ltADDRESSgt
• ltINSTITUTEgt IISc lt/INSTITUTEgt
• ltCITYgt Bangalore lt/CITYgt
• lt/ADDRESSgt
• lt/STUDENTgt
• ltSTUDENTgt
• ltNAMEgt Charuta lt/NAMEgt
• ltADDRESSgt IIT, Mumbai lt/ADDRESSgt
• lt/STUDENTgt

6
Example (contd.)
7
XML to Relational

• 1. STUDENT1 (KEY, NAME, INSTITUTE, CITY)
• STUDENT2 (KEY, NAME, ADDRESS)

STUDENT
NAME
ADDRESS
IIT, Mumbai
Charuta
8
XML to Relational (contd.)

• 2. STUDENT (KEY, NAME)
• ADDRESS1 (KEY, INSTITUTE, CITY)
• ADDRESS2 (KEY, ADDRESS)
• 3. STUDENT (KEY, NAME, INSTITUTE, CITY, ADDRESS)

STUDENT
NAME
ADDRESS
IIT, Mumbai
Charuta
9
STORED (Deutsch et al.)

• A query language used to define mappings
• Q FROM STUDENT X
• NAME N,
• ADDRESS INSTITUTE I, CITY C
• STORE STUDENT(X, N, I, C)

10
STORED (contd.)

• Q FROM STUDENT X
• NAME N,
• ADDRESS INSTITUTE I, CITY C
• STORE STUDENT(X, N, I, C)
• Mappings generated through data-mining algorithms
• Non-conforming data stored in overflow graphs

11
XML to OO (Christophides et al.)

• An OO schema is derived based on the DTD
• lt!ELEMENT STUDENT (NAME, ADDRESS)gt
• lt!ELEMENT NAME (CDATA)gt
• lt!ELEMENT ADDRESS (INSTITUTE, CITY)gt
• lt!ELEMENT INSTITUTE (CDATA)gt
• lt!ELEMENT CITY (CDATA)gt

12
XML to OO (contd.)
class STUDENT public type tuple ( name
NAME, address ADDRESS ) class ADDRESS
public type tuple ( institute INSTITUTE,
city CITY ) class NAME inherit Text class
ADDRESS inherit Text class INSTITUTE inherit
Text class CITY inherit Text
13
Natix (U. Mannheim)

• A basic record manager is used
• Each node or set of nodes is stored in a record
• Splitting strategies on the tree can be employed
  to optimize query processing

14
Natix (contd.)
15
Summary of issues involved

• Determining the best mapping
• Space occupied
• Data fragmentation
• Support for overflow data
• Lossless?
• Translation of XML queries to DB queries
• Reconstruction of XML documents from the DB