Databases Reference
In-Depth Information
query processing in this case, it is necessary to provide the optimizer with
adequate information, for example, some knowledge of how it can estimate
the cost of evaluating newly defined predicates.
For a typical example of the aforementioned domino effect, consider
concurrency control for access paths. Many DBMSs use specialized concur-
rency control protocols on indexes to prevent unnecessary locking conflicts
[31, 32], which otherwise would increase lock contention and decrease trans-
action throughput. Therefore, whenever a new index is introduced, concur-
rency control (for this new index type) also should be adopted, which is not
possible in current systems.
12.4.2
Middleware DBMS
This type of CDBMS aims at integrating external data stores into DB sys-
tems without requiring the data to be moved under control of the DBMS.
Still, users should be provided with an integrated view, that is, queries and
updates on the data source should be possible, and the disparate data formats
should be resolved into a homogeneous form.
The underlying problem in that respect is that the DBMS needs to
understand the data formats and the functions of each data source. Two
extreme alternatives exist to tackle the problem. In one, information about
the data sources interfaces is hardwired into the (integrating) DBMS. The
realm of integratable external data stores is thus restricted, and the DBMS
needs to be extended for each specific type of data store. In the other
extreme, a common data model, query language, or interface to external data
stores is set as a prerequisite (e.g., SQL). Each data store that does not imple-
ment SQL right away would have to be extended to do so.
The solution that helps to overcome the intrinsic problems of both
approaches lies in the introduction of additional components. In a nutshell,
one component is pushed between the DBMS and each data source. The
components serve to homogenize differences in formats and functionality
from the DBMSs point of view. From the data sources perspective, they
level the different data source capabilities into a common basis. Thus, each
component mediates between the data sources and the DBMS orfrom the
DBMSs point of viewwraps the data source.
The first prerequisite of this approach is a common data abstraction
(e.g., objects). Second, the mediation components offer a common interface.
This interface is used by the DBMS to request data from the data sources.
Each component should at least support the minimal interface, such as
scanning over a collection of data entities. Depending on the data source
