DB2 UDB for z/OS Version 8 Performance Topics - IBM Redbooks

Rules for query rewrite
If the query references tables that are also referenced in MQTs, the optimizer can decide to
select the qualified MQT that gives the best performance for the query.
Figure 3-7 summarizes the checking that is performed for a query, if automatic query rewrite
is generally enabled in your DB2 subsystem at installation time.
The automatic query rewrite is supported for dynamically prepared queries that are read-only
and have some other restrictions.
The automatic query rewrite is a general process that can use a materialized query table M if
the fullselect F that makes up the MQT has common source tables as the submitted query Q
that is being optimized by DB2.
Other conditions are checked during table mapping to verify if the result of the submitted
query can be derived from or can directly use the result of one or more materialized query
tables.
During a predicate matching process, the predicates in the materialized query table fullselect
are compared, one by one, to the predicates in the query Q.
Diagrammatic Overview of Automatic Query Rewrite
Table mapping
Common table: table in both Q and F
Residual table: table in Q only
Extra table: table in F only
Rejoin table: common table joins with
M to derive non-key columns
Predicate matching
Join predicates: exact match in both Q and M
Local predicates: P in Q subsumes P in M
C > 5 subsumes C > 0;
C in ('A', 'B') subsumes C in ('A', 'B', 'C')
In other words, M contains data that Q needs.
Figure 3-7 Rules for query rewrite
Table
mapping
Restrictions on Query
Subselect that can be rewritten:
Read-only query
Contains base tables & table
functions, not outer joins only
No ROWID, LOB types
Heuristics of Selections
Match with no regrouping, no residual
joins, and no rejoins.
Match with no regrouping and no
residual joins
Match with the largest reduction ratio:
|T1|*...*|Tn|/|M|, where T1, ..., Tn are
base tables in F
Predicate
matching
Grouping
matching
Expression
derivation
Restrictions on Fullselect
Base tables & table functions
Single SELECT after view and table
expression merge
No ROWID, LOB types
No outer joins
.
Grouping matching
Functional dependency: column K determines C
Matching requirement: Grouping column in F
determines grouping column in Q
No regrouping requirement: Grouping column in
Q determines grouping column in F
Expression derivation
Column equivalence: T1.C1=T2.C1, T2.C1 can be used
for T2.C1
Arithmetic expressions: C*(A+B) can be derived from
B+A and C
Scalar functions: substr(C, 5,10) can be derived from C
Set functions:
AVG(C) = SUM(C)/COUNT(*) if C NOT NULL
VAR(C) = ...
Rejoin table columns: From key column T.K derive
non-key column T.C
If there are any predicates that are in the materialized query table subselect, but are not in the
query, then it can be assumed that these predicates may have resulted in discarded rows as
the materialized query table was refreshed, and that any rewritten query that makes use of
the materialized query table does not give the correct results.
The GROUP BY clauses, if any, are compared during grouping matching, to determine if a
query GROUP BY clause results in a subset (not necessarily proper) of the rows that are in
Chapter 3. SQL performance 43