Nowadays, business processes evolved to networked workflows that are complex and executed in parallel with
little human involvement to meet the needs of today's agile and adaptive business. Contemporary business
requirements yaw for agility, flexibility and service orientation. A simplified summarization of this widely
discussed and necessary business trend can be reduced to the demand, that today's businesses have to adapt
their processes and organizations faster than their competitors. Busi- ness organizations that are able to
handle critical business events faster than their competitors will end up us winners in today's globalized
and fast business.
The pillars of such business models are loosely coupled, distributed and service- or event driven-oriented
systems that generate huge amounts of events at various granularity levels. The lack of tracking those events
and maintaining the causal relationships and traceability between those events, as well as aggregating them
to high level events or correlating them, is a problem that is currently investigated by many research
groups.
Event-based systems are increasingly gaining a widespread attention for such classes of problems, that
require integration with loosely coupled and distributed systems for time-critical business solutions. The
field of event-based or event-processing systems is a quite young area of research and is mainly influenced
by the publish-subscribe paradigm and relational database and later on by Active- and Zerolatency data
warehousing.
A promising solution for these problems is Complex Event Processing (CEP). The term of Complex Event
Processing (CEP) was first introduced by David Luckham in his book The Power of Events and defines a set of
technologies to process large amounts of events, utilizing them to monitor, steer and optimize the business
in real time. A CEP system continuously processes and integrates the data included in events without any
batch processes for extracting and loading data from different sources and storing it to a data warehouse for
further processing or analysis. CEP solutions capture events from different sources, with different time
order and take events with various relationships between eachother into account.
The contributions of this dissertation are settled in the research area of event processing systems with a
special focus on CEP and Event Processing- and Query Languages.
Event-Based Component Model: The presented work introduces an event component model that puts event-based
systems into a broader context of event processing, defines the boundaries of such systems and the scope of
event-driven components. In particular the introduced model decouples event-based system completely from the
communication infrastructure, which offers the advantage that the capabilities of an event processing realm
are not constrained by the underlying communication technology.
Event-Processing Models: Several new concepts of event models are introduced as their design strongly
constraints the capabilities of event processing query languages. Further they have a ma jor impact on the
flexibility and usability of event-based systems. A special attention in this thesis was also set on
event-driven sense and respond rules, which can be used to model trees with event actions within the event
processing model.
Event-Base: This thesis introduces the design and the concepts of the Event-Base which is an extension of
the event processing system SARI. SARI allows to observe relevant business events to identify exceptional
situations, indicates opportunities or problems combined with low latency times in decision making for
supportive or counter measures. The Event-Base, on the other hand, provides an efficient up-to-date
operational storage together with retrieval mechanisms for business events for analytical as well as
operational purposes without the costly data staging processes known from established data warehousing
solution.
SARI-SQL Query Language: The ma jor contribution is the introduction of the syntax, semantics and the
evaluation of the event query language SARI-SQL and its sub-language the EAExpression. SARI-SQL can be
allocated to the group of domain-specific languages. The query language allows to retrieve near real-time
events and create conjunctions with historical events, metrics and scores. Furthermore, it is in contrast
to Event Clouds indexing approach a formally structured solution that extends ANSI-SQL. SARI-SQL creates an
abstraction of the event type model by encapsulating a lot of overhead and creating an abstraction layer over
events and their internal data structures. The user of this language can concentrate on only expressing the
required results instead of putting effort into making the "things run". As a consequence it allows domain
experts to easily gain insights due to the level of abstraction of the specific problem domain. So for
instance it allows retrieving events according to several constraints and access event correlations.
The results of this dissertation provide the research community as well as interested parties with a
generic component model for event-based systems. The introduced model has been successfully evaluated through
the implementation of the Event-Base. SARI-SQL is an integral part of the Event-Base and its implementation
was a ma jor challenge in terms retrieving events and their correlations in a reasonable time. The presented
work set a special focus on a clean and expressive language design in order to encapsulate all the
event-related entities. Furthermore an emphasis was set on an efficient design of the query preparation and
evaluation architecture that allows attaching different query optimizer strategies. With the introduced
optimizer strategies the performance of queries on single-value types (which applies in 80% of the cases) is
directly correlating with the underlying RDBMS performance constraints and thus creates only a small
overhead.
The future work on SARI-SQL includes efforts in optimizing the strategies of handling nested attribute types
of events. This includes query analysis procedures and execution planning strategies in order to reduce the
number of in-memory post-evaluation operations.
The presented work is part of a long-term research effort aiming at designing and developing a comprehensive
event analysis toolset that allows users to query and analyze large repositories of real-time and historical
events from various sources. In addition the goal is to consolidate and create a rich unified event model for
event-based systems which can be supported by a wide range of event-based systems. A key focus of future
research is also set on the aspect of the visualization of events with respect to their temporal occurrence,
their correlation with other events, and event clusters.
