XML et' al vs Tables

1
XML et. al vs Tables

• Michael Stonebraker
• Adjunct Professor
• Massachusetts Institute of Technology
• (stonebraker_at_lcs.mit.edu)

2
Tables

• Good for structured data
• remember Codds laws
• Schema first
• have to think about your data up front
• Data integrity key
• not loosey-goosey semantics

3
XML

• Good for documents
• Bad for data (Codds laws)
• Schema later ok
• Loosey-goosey ok

4
Applicability
lots
Tables
data volume
XML
little
little
lots
structure
5
Why?

• Low end content uses HTML -gt XML
• High end content uses DBMS
• invariably structured
• High end business data processing uses DBMS
• Low end may well use XML
• because it is easier

6
Which is More Important?

• Internet has generated lots of HTML content (on
  the left)
• As well as lots of click stream data (on the
  right)
• Which is more important depends on metric
• bytes represented?
• bytes monetary value of a byte?
• market value?

7
Nothing is New

• Document systems have never used data bases
• IR community versus DBMS community
• XML will allow better IR
• Outer fringes have never used data bases
• in the 1970s it was airline reservation systems
• in the 1990s it was web crawlers

8
But DBMS is Good

• Makes you think about you data up front
• good discipline (used by business data processing
  folks)
• registers meta data for downstream data
  sharability
• Transactions, recovery, etc. are good

9
Battleground Between The Two Approaches

• Corporate media asset management
• Scientific data

10
How Can DBMS Do Better?

• Become easier to use
• Better tools

11
My Main Fear

• Two fiefdoms are incompatible
• shades of Corba and Com
• cross fiefdom joins may not work well
• DBMS will export XML
• no guarantee that client wont update it and hand
  it back
• horrible view update problem.