Database Reference
In-Depth Information
Thus, after we gather enough evidence about the usefulness of a given
candidate index, we create supporting statistics to have more accurate
information in the near future.
10.3 Summary
Motivated by complex applications with unexpected workload characteristics,
new approaches to the physical database design problem are required. To
address such scenarios, we showed progressively more dynamic solutions:
•
Simple adaptations of oine tools that periodically tune workloads and
automatically deploy the resulting configurations
•
Alerting mechanisms that very eciently can determine whether a tun-
ing session would be worthwhile and therefore would avoid wasting re-
sources in unneeded but expensive tuning sessions
•
Continuous tuning, which incrementally maintains the physical design
of a database reacting to changes in workload patterns and takes into
account the cost to transition between physical design configurations.
10.4 Additional Reading
Bruno and Chaudhuri introduce the alerting mechanism described in Section
10.1 and give more details on how to handle
UPDATE
queries and views and
how to obtain upper bounds on improvement (this effectively reduces the
chances of false negatives, by bounding the best possible outcome of a com-
prehensive tuning tool).
3
,
4
Most of the content of Section 10.2 on continuous
physical design tuning is taken from the work of Bruno and Chaudhuri. Some
recent work in the literature describes prototypes that address online phys-
ical design tuning,
2
,
5
,
6
and discusses how to benchmark different proposals.
7
A very recent thesis discusses online tuning in great detail, and presents a
principled approach based on index interactions that refines the approach de-
scribed in this chapter.
8
A slightly different but related approach is an oine
technique that finds the optimal physical schedule considering the workload
as a sequence,
1
which helps developers writing code that takes into account
physical design changes.
Search WWH ::
Custom Search