Inktomi Database (simplified)

Two kinds of tables:

• Relation, w, for every word: w[doc id, score, position info]

  • About 10M words,
  • Average of 1000 rows/word, but very long tail
    

• Relation, D, for every document: D[id, url, abstract, …]

  • Currently 55M documents

All tables fragmented horizontally by id

• Ids are unique (no duplicates)
• Complete and disjoint fragmentation

Previous slide

Next slide

Back to first slide

View graphic version