Database Reference
In-Depth Information
Figure 2. Examples of benefits on the use of smaller facts table fragments at the global level
In the case of large dimension tables' fragmen-
tation, some data replication may also occur. For
instance, let's consider a grid-DW of a nation-wide
retail store. There are several sites participating
in such grid-DW, each one at a distinct state. In
such warehouse, a (large) dimension table stores
information about customers. Such table may be
fragmented according with the location in which
the customer buys. Initially, each customer's
information would be at a single site. But when
a certain customer travels (or moves) to another
state and buys at stores from such state, his/her
information may also appear in the state's database.
When there is dimension table (fragment) replica-
tion at distinct sites, a replica consistency strategy
may be necessary. There are several works in the
literature about algorithms to efficiently maintain
replica consistency in distributed and grid-based
databases [e.g., (Akal et al, 2005; Breitbart et al,
1999; Chen et al, 2005)].
Hence, in the GPS, the facts table is partitioned
by the combination of the site source attribute with
the other most frequently used equi-join attributes.
Derived facts table partitioning may be used. Each
site stores the partitions from which the site is
the Data Source Site. Large dimension tables are
fragmented and small dimension tables replicated
at all sites. A fragmented
site source
dimension
table (each site storing only its own information)
should be used. Such data distribution strategy is
represented in Figure 3. Facts table's fragments are
replicated across grid's sites in order to improve
performance and availability.
The QoS-Oriented Query Scheduling
Users submit queries to the grid-based DW
considering the Logical Model. Ideally, the
physically used data distribution strategy should
be transparent to users. Hence, the first phase in
Search WWH ::
Custom Search