Databases Reference
In-Depth Information
[21] is based on the ideas coming from extensible DBMS, adopted to scal-
ability and elasticity needed by cloud environments. They propose Cloudy a
modular cloud storage system, based on a generic data model, and which can
be customized to meet application requirements. Cloudy offers to an applica-
tion several external interfaces (key/value, SQL, XQuery, etc.) while supporting
different storage engines (such as BerkeleyDB or in-memory hash-map). Each
Cloudy node in the system is a autonomous running process executing the en-
tire Cloudy stack. Cloudy supports several replication and partitioning schemes
for distributing the load across nodes. The main problem with such approaches
is that the addition or modification of components requires high specialist and
even programmers need to be aware of many technical details to master this
type of platform.
4.2 Transparent Access to Integrated Data Stores
It is a classical problem which has been widely addressed by the data media-
tion community [22]. Two types of components are used to allow different data
sources to be integrated and manipulated in a transparent way: a mediator and
adapters. The mediator hides the sources to applications. It implements a global
schema described in a neutral data model and including all the information
to share. There are two approaches to design the schema, global-as-view (the
global schema is defined as a view on the data sources) and local-as-view (the
data sources are defined as views on the predefined global schema). Applica-
tions issue queries on the global schema using the query language associated to
the neutral data model and the mediator rewrite these queries as sub-queries
send to the target data sources. Each data source is encapsulated by an adapter
transforming a neutral query into a (set of) data source query.
In our context, this approach is not so easy because of the important het-
erogeneity of the data stores, especially related to their data models and query
languages. [23] is a first attempt in this direction. It proposes a mediation system
aiming at supporting document and column family data stores. The relational
data model is used to define the global schema whereas NoSQL and relational
DBMSs are target data sources. A mapping language has been proposed to map
attributes of the data sources to the global schema. The mapping language sup-
ports different access paths to the same relation as well as multi-valued or nested
attributes. Query rewriting is done using a bridge query language allowing to
bridge the gap between a declarative language as SQL and the imperative lan-
guages implemented by many NoSQL data stores. This proposal is promising
even if some functionalities are lacking (no query optimization at the global
level for example).
4.3 Consistency, Scalability and Elasticity
Scalability represents a static property that specify the behavior of a system
based on a static configuration. Elasticity defines a dynamic property that
enables to scale up and down the system on demand, at runtime.
Search WWH ::
Custom Search