Clustering
  Placement of  "similar" records together.
  Critical for performance because
  Multiple records fit on one page, so a single page access can get at many
    related records. 
  Sequential I/O is much cheaper than random I/O for a disk page.


Multiple Dimensions
  A relation has multiple attributes.
  Cannot cluster records based on all of them simultaneously.
  Build a clustered index on the most "important" attribute.
  Build unclustered indices on other selected attributes.
  No indices for "less important" attributes.


Fractals
  Take a base shape and apply expansion rule repeatedly.
  Each recursion grows cure by multiplicative factor.
  Space filling curve
  Self-similarity