Database Reference
In-Depth Information
More precisely, at each node in the MVMG, are
stored its frequency, size and meantime (execu-
tion time). While in related work, the frequency
is considered at the user queries level. The first
advantage of our approach is to exhibit frequent
views even if the related query is not frequent.
This means the view is used by many different
queries. Therefore, the aim is to improve the ef-
ficiency of the view selection algorithm.
The second convenient feature appears at the
operation level since we may take the parallel-
ism into account. The MVMG framework also
allows to clean the MVMG by eliminating the
non-frequent views rather than entire queries.
Building the MVMG is the main task for setting
up the search space by identifying common sub-
expressions. This task is of importance similar to
multi-query optimization. However, it is orthogo-
nal to the view selection process itself. This paper
focuses on the view selection process.
the method, the other important argument of our
solution is reducing the complexity of the view
selection algorithm. The second argument for
considering the views to materialize query by
query relies in preserving the data independence
whenever adding a query to the view configura-
tion or removing one from it. Indeed, the side
effect on existing queries is reduced since the
view selection is applied query per query. In the
related work since the view selection strategy
consists in considering all equivalence nodes of
the multi view query graph, the impact of adding
or removing a query may lead to an important
reorganization.
Criteria of Selection
To each view v is assigned a value, called
goodness(v
), which represents the benefit of
materializing v. This value is computed by the
following formula:goodness(v) = freq(v) × cost(v)
Where freq(v) is the access frequency of view v
and cost(v) is its processing cost. Unlike related
work, our approach is based on statistic metadata.
The cost, for instance, is computed according to
the average query processing time for a given
view v:cost(v) = meantime(v)
VIEW SELECTION STRATEGY
In this section, we present our method for dynami-
cally selecting a set of views for a given query and a
predefined storage space constraint. The constraint
of view maintenance is relaxed since our method
is dynamic: we are not considering a workload of
a set of queries but a given query. However, our
method takes the reuse parameter into account in
order to reduce the view maintenance cost and
the storage space. At first, as it was described in
the previous section, the MVMG allows to detect
common views. Secondly, the reuse is also handled
by the frequency of a view.
Pre-Selection of Views per Query
At this stage, we try to avoid candidates which
do not provide additional benefit to the set of
already materialized views. A given query q is
decomposed into a set of views. The method pre-
selects a subset of views called candidate views
V
C
, from the set of views from q.Goodness(V
c
)
= ∑goodness(v, V
c
) be maximalv ∈ V
c
|V
c
| be
minimalWhere:goodness(v, V
c
) = goodness(v) if
no view in V
c
is subsumed by v 0 else
The previous formula is designed in order to
avoid the case where the view v is subsumed by
another materialized view v' because materializing
v provides no additional benefit for processing q.
Unlike other approaches, ours tends to materialize
Selection and Dynamic
Materialization
The process of view selection/replacement is
automatically triggered by the arrival of a user
query or when an expert-user decides to materi-
alize a query. Besides of the dynamic feature of
Search WWH ::
Custom Search