Database Reference
In-Depth Information
self-tuneable solution that relieves the warehouse
administrator from having to monitor and calibrate
the system constantly.
The view pool is the information repository
used for storing materialized results as multidi-
mensional data structures. For each query that
arrives at the system, first the fragment locator
determine if it can be answered from the already
materialized view by linearly searching the pool in
the aide of dictionary index. In the case of failure
in finding any already materialized view, the query
will be answered conventionally from the base
tables. Either-way, after the result is computed
and given to the user, it is tested by the Admission
Control Entity which decides whether or not it is
beneficial to store it in the Pool.
They provide a replacement policy based on the
goodness of a view. Different kinds of goodness
have been experimented: last date of access, use
frequency, size and cost of re-computing a view.
However, Dynamat's approach dealt with query
per query and considers the materialization at
the final result of a query. Hence, it did not use
any framework to detect common views between
queries for reuse purpose.
More recently, a formal study of the view
selection problem focusing on its complexity has
been done in (Chirkova, 2001). It shows notably
that the cost model is a parameter of importance
in the view selection setting.
In paper (Ghozzi, 2003) the authors studied
the impact of the constraints on the manipulation
of dimensional data. More precisely, it analyzed
the repercussions of the constraint over the di-
mensional operators and the query optimization
by exploiting the semantic constraints for the
selection of materialized views. This approach
uses a dimensional lattice which is based on
the structure of the hierarchies and which also
integrates the constraints. Thereafter, the views
which do not satisfy the constraints are removed.
The author did not propose an algorithm or view
selection method, which could optimize the query
processing indeed.
Another dynamic approach presented in
(Schnaitter, 2006) provides the solution for
materializing the indexes which can be seen
as a special case of the materialized view. The
authors introduce a novel self-tuning framework
that continuously monitors the incoming queries
and adjusts the system configuration in order to
maximize query performance. They call it CoLT
(Continuous On-Line Tuning). It supports the on-
line selection of an effective set of indices for a
relational query by the following steps:
1.
CoLT builds a model of the current
workload based on the incoming flow of
queries.
2.
It estimates the respective gains of differ-
ent candidate indexes.
3.
It selects those that would provide the best
performance for the observed workload
within the available space constraint.
Thus, the system performs continuous profil-
ing and reorganization in order to match the most
recent traits of the query workload.
Paper (Zhou, 2007) presents another idea to
add the dynamicity to their approach. Unlike
other dynamic approaches, the dynamicity is not
applied for the selection of the view but instead
they suppose that the materialized view is already
selected. However, the dynamicity is applied at
the view extent level. Therefore, the algorithm
selectively materializes only the most frequently
accessed tuples of the view, and these tuples can
be dynamically changed.
When any query arrives to the system two step
of testing are required in order to decide if it is pos-
sible to answer the query from the materialized part
of the view. The first test could be accomplished
at the optimization time as for regular view that
tests the containment of the query in the view.
However, the other can be done only at execution
time which called guard condition. This condition
checks whether one or a few covering values exist
in the control table. The construction of the guard
Search WWH ::
Custom Search