From: Atri Rudra (atri@cs.washington.edu)
Date: Sun May 23 2004 - 22:32:08 PDT
Access Path Selection in a Relational Database Management System
-----------------------------------------------------------------
P. G. Selinger et al.
This paper describes the query optimizer in System R. The basic idea is to
choose among different access path selection which roughly means selecting
the order of logical operators and choosing physical operators to
implement them. In other words, one searches for the optimal plan from the
search space of "feasible" plans where the optimal is defined by some cost
function which requires defining the parameters (System R maintains
statistics about these parameters) and the functions themselves.
These are my comments/questions:
(i) The optimizer in System R checks for semantic errors and type
compatibility: can't this be done at the parsing stage ?
(ii) The paper talks about interesting orders which have impacts on
queries having group by/order by clauses and join results. Can one use
some informations about memory layout to define some interesting order ?
(iii) The paper says that the merging scan joins though applicable for all
kinds join predicates is used for equijoin predicates: is this because of
the fact that < or > is not commutative ?
(iv) It is interesting that the Dynamic Programming method is not
mentioned by that phrase: was it not common when the paper was written ?
(v) I read the Chaudhari paper before this and this read turned out to be
a repeat read for most parts.
This archive was generated by hypermail 2.1.6 : Sun May 23 2004 - 22:32:08 PDT