Data exchange by example - Foundations of Data Exchange - page 11

desired answer. The complexity of this task will also depend on a class of queries to
which Q belongs. We shall see that for some classes, constructing Q is easy, while for
others Q must come from languages much more expressive (and harder to evaluate)
than SQL, and for some, it may not even exist.
1.3 Data exchange vs data integration
We comment briefly on the connections between data exchange and the closely related field
of data integration. In a typical scenario of (virtual) data integration we have one or more
local sources, as well as a global schema, and a specification of the relationship between
the local schemas and the global schema, i.e., a mapping. The data resides at the local level,
but the user wants to query a virtual database of the global schema. Thus, superficially, a
data integration system can be seen as a data exchange system: we view all local sources
as a single database of a local (source) schema, and, under a given mapping, want to query
a database of global schema (target). Moreover, as in both cases the target, or the global
schema database, is not unique, the semantics of query answering is provided by certain
answers.
The key difference between data exchange and virtual data integration is in the treatment
of target (global schema) data, and in the type of data used for query answering. In virtual
data integration, we do not materialize a database of the global schema. Instead, we rewrite
a given query Q over the global schema into queries over the sources so that their results
would either compute, or approximate in the best way, the answer to Q . That is, the target
is not materialized, and source data is used for query answering.
In data exchange, on the other hand, we do materialize the target. Of course there could
be many target instances, but we choose to materialize one of them, and then use it to
answer queries Q . We may choose to run the same query Q , or rewrite it into a different
one, but we do run queries on the materialized target. That is, in data exchange, the target
is materialized, and it is that materialized target data that is used for query answering.