View Management Techniques and Their Application to Data Stream Management - Evolving Application Domains of Data Warehousing and Mining - page 102

mantics (Jagadish et al., 1995; Liu et al., 1999)
were basically systems that defined a continuous
query as a view and periodically re-evaluated the
query by updating the view.
However, there are also fundamental differ-
ences between incremental view maintenance and
continuous query evaluation. In contrast to view
maintenance, periodical updates processed in a
batch-like mode are not an option in stream data
management. Instead, DSMSs adopt a pipelined
approach to query evaluation. Insertions to and
deletions from the instantaneous relations are
handled by the query operators as soon as they
occur (albeit operators are scheduled following
some policy). Thus, it is not necessary to poll the
base relations for changes.
As discussed in Section 3.2 view maintenance
techniques have been implemented for different
data models and with different query languages.
Predominantly, data streams are based on the
relational model and can be continuously queried
with extensions of SQL. There exist also systems
which offer continuous queries for XML data,
such as the NiagaraCQ system using XML-QL
(Chen et al., 2000).
In view maintenance, self-maintainability is
desired, i.e., maintenance of the view without
access to the base relations. These base relations
correspond to the instantaneous relations speci-
fied by the streams and their window definitions
in a continuous query. Although, it is of course
desirable to evaluate continuous operators in-
crementally without accessing the underlying
instantaneous relation, it is often (as in the case of
holistic aggregation operators) inevitable to scan
the whole window to evaluate an operator.
Also, in view maintenance any tuple can be
deleted from the underlying base relations. In
DSMS, instantaneous (base) relations are append-
only. Though, if a window is defined, tuples are
removed from the instantaneous relation either
time-based or order-based. They are only removed
in the same order in which they were inserted. It

is not possible to remove a tuple from the stream.
Eventually, any tuple will flow out of a bounded
window and hence be not further considered for
evaluation.
Thus, with respect to the information dimen-
sion in the classification of view maintenance
approaches, answering continuous queries can-
not be restricted to self-maintainability. On the
other hand, the strict order of tuple removal
from the instantaneous relation may offer some
advantages.
Data streams produce a vast amount of data
which sets new challenges for the (real-time)
processing of data. Efficient techniques for rapid
processing of incoming tuples are required. The
rate of changes to the instantaneous relations is usu-
ally considerably higher than in view maintenance.
This causes new challenges for efficient evaluation
of operators. Highly optimized operators such as
the state-slice join described in section 6.3 allow
fast, and truly continuous query evaluation.
However, in general the high data arrival
rates in DSMS require query answers to be ap-
proximated. Thus, as opposed to view maintenance
which usually requires exact answers, results to
continuous queries are usually only approximated.
This approximation is implemented by window
models as described in Section 5.2 and synopsis
depicted in Section 5.3. The timeliness is an in-
herent property of data streams. As streams are
continuously flowing in the instantaneous relations
are updated as window operators slide on and
therefore guarantee an immediate incorporation
of changes.
Instead of considering stream data management
as a special case of view maintenance, it may be
beneficial to swap sides and regard view main-
tenance as a special case of stream data manage-
ment. This could allow to leverage the efficient
data stream query operators for fast incremental
view maintenance. Such an approach would be
very similar to real-time data warehousing or
online view maintenance (Quass and Widom,