Schema Evolution in Wikipedia

1
Schema Evolution in Wikipedia
Toward a Web Information System Benchmark
Carlo A. Curino Hyun J. Moon Letizia Tanca Carlo
Zaniolo
2
Motivations

• Understand the role of the Schema Evolution (SE)
  in Web Information Systems (WIS) our guess was
• on the web everything evolves faster -gt SE
  should be relevant!
• Compare the evolution in Traditional IS (known)
  and WIS
• Obtain an in-depth understanding of Wikipedia
  DB backend
• Lay the foundations of a Benchmark for Schema
  Evolution
• Why MediaWiki (software platform behind
  wikipedia)?
• Popular (used by gt30.000 websites including
  Wikipedia)
• open-source and well-documented software
• Wikipedia DATA and QUERIES also under
  open-source license

3
What we did

• We developed a tool-suite to analyze Web
  Information System DB backends
• We collect and dissect MediaWiki schema history
  (170 schema versions in 4.5 years)
• We release tool-suite and data as a first step
  towards a Benchmark for Schema Evolution

4
MediaWiki Architecture

• Classical Web architecture based on Linux,
  Apache, MySQL, PHP (LAMP)
• Big scalability issues
• Wikipedia is one of the 10 most popular websites
  in the WWW (about 29k requests/sec in average,
  peaks up to 85k requests/sec)
• Several Layers of caching (both explicit and
  not, at DBMS and WS level)
• According to the developers DBMS performance are
  the major bottleneck, DB size gt 700Gb not
  considering the multimedia content!!
• plus poor load partitioning (one-language per
  server)!

5
The Schema

• Tables can be grouped in
• article and content
• links and structure
• users and permissions
• performance and caching
• statistics and special features
• history and archival (represent a big portion of
  the schema 1/3 they dont know it but they need
  a temporal DB!)

6
Basic Statistics 1

• Schema Evolution
• 170 versions in 4.5 years
• almost 250 increase

7
Basic Statistics 2

• More frequent schema changes far away from
  releases
• Schema Elements Lifetime
• a group of stable relations
• young tables and columns

8
Type of Changes

• NOTE it doesnt adds up to 100 since several
  changes might coexist in an evolution step
• total lack of integrity constraints (a part from
  primary keys)!!

9
Type of Changes

• NOTE simple schema modifications are the most
  common

10
Schema Changes per Version

• NOTEversion 41-42 represents a MAJOR evolution
  step where article versioning management is
  heavily modified!!

11
Impact on the Applications

• NOTE over 4000 queries from which we extract
  75 templates

12
Wikipedia Profiler Queries

• NOTE 500 most common templates out of 2k
  extracted from over 780 millions query instances
  from the On-Line Wikipedia Profiler
  http//noc.wikimedia.org/cgi-bin/report.py

13
Traditional vs Web IS

• Comparing our results with existing analysis for
  Traditional IS
• WIS evolve faster 38 (w.r.t. Sjoberg) and 539
  (w.r.t. Marche)
• Collaborative WISs embrace information sharing,
  thus we got way better data 170 versions vs 2
  and 9 respectively. And we can share them
  (benchmark).
• More in-depth analysis by means of SMOs

14
Towards a Unified Benchmark

• We share the schema history we collect, the
  analysis data (raw and stats), the queries, the
  tool-suite
• http//yellowstone.cs.ucla.edu/schema-evolution/in
  dex.php
• Goal create a benchmark for schema evolution
  and in general a standard relational DB dataset

15
Conclusion

• So far we
• Create a tool-suite for schema evolution
  analysis
• Dissect Wikipedia Schema Evolution history
• Establish the core of a DB Schema Evolution
  Benchmark (released)
• We developed tools to support Graceful Schema
  Evolution (PRISM)
• We plan to
• Extend the analysis to several other Open-Source
  WIS (Joomla!, TikiWiki, Slashcode, Zen-Cart)
• Extend the analysis towards public scientific DB
  (Genome, HGVS)
• Involve other research groups to define a
  commonly-agreed Benchmark
• Improve the tool-suite and integrate it in PRISM