Pears

1
Pears

• Yet Another Database Fruit

Ralph LeVan
http//www.oclc.org/levan/docs/pearsoverview.ppt
2
Pears

• Replaces Newton (and SMART)
• Easier to Use
• More Functionality
• More Reliability

3
What We Are Replacing?

• Newton at the heart of critical productions
  systems
• Used by most projects in the Office of Research
• SMART critical to Mantis (and hence CORC) but
  can not be used commercially

4
Easier to Use Than Newton

• Fewer Files
• Files Grow Automatically
• Fewer Programs

5
Five Files Become One
Newton
PDB
6
Automatic File Growth

• Newton files are preconfigured for size. (IBM
  Mainframe file model)
• Pears databases grow automatically. (Standard
  UNIX file model)

7
Five Newton Programs...
Initdb
8
Become One Pears Program
Bartlett
Pears
9
More Functionality

• Unicode.
• User Customizable Indexing.
• Relevance Ranking.
• Online Update
• Embeddability

10
Online Update

• Newton databases could not be searched and
  updated at the same time.
• Pears databases implement an optimistic collision
  model instead of record locking.

11
Embeddability
CORC
Z39.50 Server
Z39.50 Protocol
Pears
Newton Database
Newton Light
Pears
Authority
12
More Reliable

• Java environment catches many errors.
• Methods throw exceptions which must be handled
  somewhere.
• Updates fail safely

13
New File architecture lets updates fail safely

• Changes to the database are written to a separate
  file and not committed until all changes have
  completed successfully.
• Changes at the end of the file are committed
  first.

14
Logical File Architecture
Abercrombie
10, 7, 5
Document 7 Abercrombie Fitch A success story
15
File Class Architecture
RandomAccessFile
16
System Architecture
17
Enhancements over Newton

• Unlimited Number of Indexes
• Index Nodes Binary-Searchable
• Short Postings Lists in Index
• Postings List Maps arbitrarily large
• Variable Length Integers

18
Conversion Programs Gone!

• Use RecordHandlers Instead!
• Can already read USMARC, SGML/XML, BER, .DB and
  .PDB
• Support user defined filters
• RecordHandler class still usable as conversion
  program

19
DBDESC Language Gone!

• Use dbdesc.ini Instead!
• DB
• blocksize16384
• InputRecordTypeSGML
• Nameddc21
• RecordHandler
• TagsFileddc21.tags

20
DBDESC (cont.)
BasicIndex index1 tagpath25/1 tagpath28/1 r
outineORG.oclc.pears.Bartlett.SmartWords Occurenc
eRoutineORG.oclc.pears.Bartlett.wordfield stopwor
dable ...
21
Summary

• Pears has replaced SMART in CORC
• Pears is replacing Newton in OR
• Users love it!

22
For More Information

• http//www.oclc.org/levan/docs/pearsdatabasefunct
  ionaldescription.html
• mailtolevan_at_oclc.org
• documentation Real Soon Now