Access Methods
  
Fundamentals
  Scan versus Index
  In what order do you get index entries?  Clustered vs Unclustered
  You can take unclustered RIDs and sort them to get "clustering"
  Hash or sort
  Double buffering

Hashing
 Static Hashing -- overflow buckets.
 Dynamic Hashing -- with and without directory.
 If directory present, double directory when needed, grow data pages by one
  unit. 
 If directory absent, shouldn't double to grow, so use temporary overflow
  chains. 

Selection
 Express in Conjunctive Normal Form
 In a conjunction, consider selectivity of predicates - choose most selective 
  index.
 Consider next-most selective and decide whether to index.
 Disjunctions as Unions
 Worst disjunct rules.  If you have to scan one, scan all.
 "Common sub-expression" hard to do, rarely done.
  
Projection
 Trivial if no duplicate elimination.
 Sorting or hashing, with obvious early elimination of duplicates.
  
Join
- Most expensive of the basic relational operators.
- So very carefully optimized.
- Many types of joins:
    Theta Join 
    Equi-join 
    Semi-Join 
    Outer Join 
    Composition 
    Structural Join 
      
- Nested-Loop Join - with blocking, with indexing
- Sort-Merge Join  (repeated values, backtrack, band-join)
- Simple Hash Join (hash smaller relation)
- Better Hash Joins use partitioning as needed
    Issue is temp-file for unprocessed partitions versus re-read of
    processed ones. 
    Partition size hard to guess, dynamic repartitioning on partition overflow.
    Many versions:
      Partition Hash Join 
      Grace Hash Join 
      Hybrid Hash Join 
- Join Indices (Cost of maintenance)
- Issues to consider include:
    Join Selectivity (hard to estimate, but e.g. nested loops may be preferred 
    if high) 
    Skew (in data) 
    Join Order 
      
Set Operations
Use hash or sort to perform set union, intersection and difference
Aggregate Operations
  Full scan of relation
  Possibly preceded by sort or hash to support grouping
  Index can be used to avoid sort/hash.