Rrelational Roots

The State of the World
 * COBOL (and precursors) ruled!!
 * Heavy industry standardization efforts around network model. 
 * Strong devotees of the hierarchical model. 
 * Codd published paper in 1970, got Turing award in 1981. 
 * Bachman got Turing award in 1973. 

When you think "out of the box"
 Very hard to give up baggage of the past completely. For example:
 * Columns are identified by position.(Notion of column names is introduced,
   but very weakly, for "user interface")
 * Entity-relationship model casts a shadow.("relationships" relation) 
 But some fundamental ideas can be thought through much more clearly.
 Ideas in this paper that have been lost over time:
 * All rows are distinct-set semantics. 
 * A semantics driven non-algebraic definition of join, allowing the
   possibility of a "plurality" of joins.  (Notion of "point of ambiguity" --
   e.g. which supplier sourced the part used in a particular assembly). 
 Ideas in this paper that have been expanded and developed:
 * Normalization and Redundancy 
 * Consistency 

Main Point
 The case for physical independence.
 * Order independence 
 * Index independence 
 * Access Path independence 
 Defeated (partially) as DB admins carefully tune large DBMS installations.

Relational Implementations
 * System R and Ingres were the the two first decent implementations of a
   relational database. 
 * "Green-fields" design means 
   - Many novel concepts were introduced 
   - Many of which have been widely adopted 
   - Others have quickly been discarded. 
 * Even so, note the influence of circumstances. 

System R
* A lot of the ideas remain in today's systems
* PL/1 and COBOL host languages 
* VM/370 operating system 

Major Influences
* SQL, including cursors, duplicates, NULLs.
* RSS/RDS divide remains in many systems 
* Cost-based optimizer 
* Updatable single-table views 
* Begin/end xact at user level 
* Savepoints and restore 
* Catalogs as relations 
* Flexible security (GRANT/REVOKE) 
* Integrity constraints 
* Triggers 
* Clustering 
* Compiled queries 
* B-trees 
* Nested-loop & sort-merge (2-way) joins 
* Dual logs to support log failure 
 
Wrong Design Decisions
* Shadow paging
* Predicate locking 
* SQL language 
* Rejected hashing