Database Reference
In-Depth Information
time-out value
T
max
that caps the maximum execution time of a query.
Although this is a practical issue to avoid very long-running queries, it
introduces some problems in the benchmark methodology. Specifically,
it changes a posteriori the optimization function that has been agreed
upon and leveraged in tuning tools. Consider the following extreme sce-
nario, with a two-query workload that contains a light query
q
1
that ex-
ecutes in 5 seconds under the base configuration
C
0
and a heavy query
q
2
that executes in 3,600 seconds under
C
0
. Suppose that a tuner
T
1
optimizes
q
2
as much as possible at the expense of not fully optimizing
q
1
, and assume that the resulting times are
(
q
2
=1
,
900
)
, with an
overall execution time of 1,905 seconds, or a 47% improvement. A sec-
ond tuning tool,
T
2
, knowing
in advance
a given 1,800 second time-out
value, might optimize
q
1
without considering
q
2
, obtaining the following
times
q
1
=4
,
(
q
1
=1
,
q
2
=3
,
600
)
, with an overall execution time of 3,601 sec-
onds, or just 0
.
1% improvement. Considering time-outs, the results are
(
q
1
=4
,
q
2
=
T
max
)
for
T
1
vs.
(
q
1
=1
,
q
2
=
T
max
)
for
T
2
, harshly underestimat-
ing
T
1
's quality.
12.5 Summary
Although much work has been done in the context of the physical design
problem, there are still significant challenges that need to be addressed:
•
Extensions that go beyond pure
SQL
workloads
•
Other related aspects in physical design, such as defragmentation or
compression
•
Interactive physical design tuning sessions for advanced DBAs
•
Robust benchmarks of physical database design tools
12.6 Additional Reading
In this chapter we discussed several challenges in the space of physical
database design. We now complement our presentation with some references
that address some of the problems presented in the preceding sections. Our
examples in Section 12.1.2 used several domain-specific extensions to
SQL
,
which are part of the
T-SQL
specification.
8
In this chapter we mentioned
other aspects of physical database design, such as the automatic defragmenta-
tion problem,
10
different challenges for recommending compressed indexes and
Search WWH ::
Custom Search