Visibility in event based systems [Elektronische Ressource] / vorgelegt von Ludger Fiege

Ludger Fiege

Sujets

• Informatik

Informations

• Publié par: technischen_universitat_darmstadt
• Publié le: 01 janvier 2005
• Nombre de lectures: 18
• Langue: English
• Poids de l'ouvrage: 2 Mo

Extrait

Visibility in Event-Based Systems
Vom Fachbereich Informatik
der Technischen Universit at Darmstadt
genehmigte
Dissertation
zur Erlangung des akademischen Grades
Doktor-Ingenieurs (Dr.-Ing.)
vorgelegt von
Dipl.-Inform. Ludger Fiege
aus Remagen
Referenten: Prof. Alejandro P. Buchmann, Ph. D.
Prof. Dr. Mira Mezini
Datum der Einreichung: 27. Juli 2004 der mundlic hen Prufung: 22. April 2005
Darmstadt 2005, Darmst adter Dissertationen D17For Biggi,
and Franzi, Flori, and JohannaPreface
Acknowlegements
Many persons have contributed to this thesis and I want to thank them for their
support. First of all, I have to thank my advisor Prof. Alejandro Buchmann
for his demanding questions, his continuous support, and his patience. With-
out his support this work would not have been possible. My co-advisor Prof.
Mira Mezini managed to point me to the engineering issues, which considerably
strengthened the statements made in this thesis.
Current and past colleagues at database and distributed systems group at
TU Darmstadt provided a wonderful working environment. Especially Mariano
Cilia, Felix C. G artner, Gero Muhl, and Andreas Zeidler endured many discus-
sions that helped me forming my ideas, and without Gero the whole Rebeca
project wouldn’t have started at all.
Last but not least, I have to thank my family for all their patience and
support. And it is always good to get a reminder once in a while where real life
takes place.
Publications
Parts of this thesis are based on previous publications. Excerpts of Chapter 2
on the formal speci cation of event-based systems are published as joint work
with Gero Muhl and Felix C. G artner [113, 114, 115]. The discussion about
content-based lters and routing in the Rebeca noti cations service is based on
joint work with Gero Muhl [214, 216]. The basis for Chapter 3 was published
together with Mira Mezini, Gero Muhl, Felix C. G artner, and Alejandro P.
Buchmann [110, 111, 115]. Some of the routing and security issues are published
in [117]. The disussion of scalability in Chapter 5 is motivated by joint work
with Gero Muhl, Felix C. G artner, and Alejandro P. Buchmann [217].
A number of additional publications have been published during the time
this thesis was prepared [45, 75, 105, 107, 108, 109, 215, 272, 290, 291, 305], and
workshop proceedings were edited that are related to this thesis [19] and to the
discussion in Chapter 2 [46, 116].
iiiAbstract
In modern IT systems, growing interconnectivity and continuous change demand
a loose coupling of the participating components and services to facilitate sys-
tem evolution. The paradigm of event-based computing and publish/subscribe
communication provide the exibility and scalability that is required in many
application domains. Examples include mobile systems, monitoring, reactive
system construction, and application integration.
In event-based systems producers and consumers of data interact indirectly.
Producers publish noti cations about events they have observed. Consumers
announce their interest in certain kinds of noti cations by issuing subscriptions.
An intermediate noti cation service conveys to those consumers
having a matching subscription. Hence, the data o w in event-based systems is
determined implicitly by set of all noti cations and subscriptions.
This decoupling of producers and consumers is one of the main advantages of
this paradigm, but in this generic form it also implies some downsides. There is
no way to di eren tiate application-level components or to tailor the functionality
of publish/subscribe services. The implied assumption is that such systems
are homogeneous. However, open systems are heterogeneous and will not only
require the middleware to accommodate to di eren t data formats but also to
combine di eren t noti cation techniques. A one-size- ts-all implementation of
the noti cation service is obviously not possible.
Unfortunately, current research and available products are primarily focus-
ing on scalability issues in terms of communication e ciency and system size,
whereas problems of system engineering and management are often neglected.
E cien t mechanisms to structure event-based systems are a missing prerequisite
for supporting engineering, management, and heterogeneity. Like in the early
stages of programming language evolution, current event-based systems are typ-
ically characterized by a at design space, with no structure and global variables
only. While producers and consumers are designed and implemented as usual,
there is no support for the role of an administrator, which is responsible for
orchestrating application components and middleware services.
This thesis identi es the visibility of components and noti cations as the
underlying objective that must be achieved for any form of interaction con-
trol. Scoping is introduced as tool for administrators to control visibility and
thus communication. A scope bundles and identi es a group of producers and
iiiconsumers. Noti cations published in one scope are, at rst, invisible in other
scopes. Scopes encapsulate the composed producers and consumers and hide the
internal structure. They localize the relationships between components outside
of the components themselves.
In order to communicate with the remaining system input and output in-
terfaces are assigned to scopes. They determine which internal noti cations are
forwarded to the outside, and vice versa. Scopes thus act as a single compo-
nent to the outside system. Furthermore, they can be recursively arranged as
members of other, higher-level scopes. Scopes and components can be
according to a variety of objectives, and at di eren t levels of abstraction. On
one hand, scopes represent common characteristics of the grouped components,
like proximity to a designated location based on geographical coordinates or
IP network topology. On the other hand, scopes delimit components of other
groups that belong to di eren t administrative domains, use other data models
for noti cations, or do not have the required security credentials. In the general
case, system engineers and administrators decide how the system is logically and
physically subdivided|scoping is the tool o ered to them by this thesis.
Once the structure of an event-based systems is identi ed, scoped noti cation
delivery can be customized. Transmission policies are associated with scopes to
re ne the rules where matching noti cations are delivered, e.g., to forward a
noti cation only to one out of a set of similar consumers in a scope. Noti c a-
tion mappings transform between di eren t data formats used for internal and
external noti cations. Together with scope interfaces this allows for a controlled
integration of independently developed and deployed pub/sub applications.
Di eren t aspects of controlling visibility of events are tackled in a large num-
ber of existing products and research contributions. However, so far an approach
orthogonal to the other aspects of publish/subscribe was missing. Scopes pro-
vide the means by which system administrators and application developers can
con gure an event-based system. Scopes o er an abstraction to reify struc-
ture and to bind organization and control of routing algorithms, heterogeneity
support, and administrative tasks to the application structure. By providing a
well-de ned point of interconnection, scopes may not only delimit the distribu-
tion of noti cations within a given service implementation but also act as bridges
between di eren t implementations. They delimit application functionality and
contexts, controlling side e ects and interaction. This is of particular impor-
tance as platforms of the future must be con gurable not only at deployment
time but also once an application is in operation.
This thesis investigates the scoping concept and its extensions in three steps.
Based on a speci cation of publish/subscribe semantics, scoping is formally de-
ned using a language adapted from temporal logic. Second, a variety of imple-
mentation strategies are compared that di er in the amount of control exerted
on communication and the type of underlying communication mechanisms, rang-
ing from point-to-point messaging to multicast, from remote procedure calls to
database storage. Finally, a prototype implementation as part of the Rebeca
distributed noti cation service is described.
ivZusammenfassung
Ereignisbasierte Softwarearchitekturen sind wegen ihrer inharenten Vorteile zu
einem wesentlichen Merkmal gro er verteilter Systeme geworden. Die lose Kopp-
lung der beteiligten Komponenten erlaubt es, autonome, heterogene Systemteile
leicht zu integrieren und die Entwicklungsfahigkeit und Skalierbarkeit der ent-
stehenden, komplexen Systeme zu steigern. Es wird zunehmend deutlich, dass
traditionelle Client/Server Ansatze, basierend zum Beispiel auf Remote Pro-
cedure Calls (RPC), diese Anforderungen nur unzureichend unterstutzen. Die
ereignisbasierte Verarbeitung verspricht hier sowohl eine hohere Leistungsfahig-
keit als auch exiblere Architekturen, die sich an veranderlic he Anforderungen
anpassen lassen.
In einem ereignisbasierten System kommunizieren Komponenten indem sie
Noti k ationen uber aufgetretene Ereignisse produzieren und konsumieren. Als
Ereignis wird dabei die Anderung eines im Computer