|
sl@0
|
1 |
Things that could be done to DBMS -- 12 October 1998
|
|
sl@0
|
2 |
|
|
sl@0
|
3 |
Query optimizer:
|
|
sl@0
|
4 |
|
|
sl@0
|
5 |
Optimize for clustering indexes (use S/T and G/H plans)
|
|
sl@0
|
6 |
e.g. assume Auto-increment column => clustering
|
|
sl@0
|
7 |
or measure the clustering in the stats
|
|
sl@0
|
8 |
|
|
sl@0
|
9 |
Optimize for multi-column lookup
|
|
sl@0
|
10 |
e.g. For "a=v1 and b=v2" with index on (a,b)
|
|
sl@0
|
11 |
|
|
sl@0
|
12 |
Optimize for like-predicates
|
|
sl@0
|
13 |
e.g. "x LIKE 'abc*'" => "x>='abc' AND x<'abd'" etc.
|
|
sl@0
|
14 |
|
|
sl@0
|
15 |
Query complexity-reduction by removing redundant predicates, e.g.
|
|
sl@0
|
16 |
"x>4 AND x>3" => "x>4"
|
|
sl@0
|
17 |
"x>4 OR x>3" => "x>3"
|
|
sl@0
|
18 |
"x>-1" for unsigned x => trivially true
|
|
sl@0
|
19 |
|
|
sl@0
|
20 |
Better statistical data dn guesswork
|
|
sl@0
|
21 |
|
|
sl@0
|
22 |
SQL/access plans:
|
|
sl@0
|
23 |
|
|
sl@0
|
24 |
Support "DISTINCT" in the projection stage via a Hash-system or sorting
|
|
sl@0
|
25 |
|
|
sl@0
|
26 |
|
|
sl@0
|
27 |
|