From: Danny Wyatt (danny@cs.washington.edu)
Date: Mon May 24 2004 - 09:37:44 PDT
Also like the first paper, this paper covered the same ground as the
previous few lectures. It is a good introduction to query planning and
how complex of a problem it is. I liked his presentation of optimal
query plan selection in the familiar terms of a search through a state
space with well-defined state transition operators. It seems like more
complex queries than those supported by System R have been integrated
into this overall approach by adding new state change operators based on
new logical (algebraic) equivalences that allow for new physical query
plans, e.g., the pulling out of joins from outer joins, the
generalization of "interesting order" to "physical properties", and
techniques learned from semijoins for handling multi-block queries.
Again, also like the first paper, I don't have much to comment on. This
seemed like a decent overview to me, but I'm new to query planning so I
don't know whether he's left out anything important. I was dismayed to
see that he says (section 7.4) that he has not covered is the on-line
construction of a plan as subplans execute. This was a new direction
pointed out in the Goetz Graefe paper, and I was hoping to see what's
become of it. I suppose that in the 10 years between then and now not
enough has been done with it to raise it to the level of interest
required for inclusion in a short overview.
This archive was generated by hypermail 2.1.6 : Mon May 24 2004 - 09:37:49 PDT