Systems for high performance computing are getting increasingly complex. On the one hand, the number of processors is increasing. On the other hand, the individual processors are getting more and more powerful. In recent years, the latter is to a large extent achieved by increasing the number of cores per processor. Unfortunately, scientific applications often fail to fully utilize the available computational performance. Therefore, performance analysis tools that help to localize and fix performance problems are indispensable. Large scale systems for high performance computing typically consist of multiple compute nodes that are connected via network. Performance analysis tools that analyze performance problems that arise from using multiple nodes are readily available. However, the increasing number of cores per processor that can be observed within the last decade represents a major change in the node architecture. Therefore, this work concentrates on the analysis of the node performance. The goal of this thesis is to improve the understanding of the achieved application performance on existing hardware. It can be observed that the scaling of parallel applications on multi-core processors differs significantly from the scaling on multiple processors. Therefore, the properties of shared resources in contemporary multi-core processors as well as remote accesses in multi-processor systems are investigated and their respective impact on the application performance is analyzed. As a first step, a comprehensive suite of highly optimized micro-benchmarks is developed. These benchmarks are able to determine the performance of memory accesses depending on the location and coherence state of the data. They are used to perform an in-depth analysis of the characteristics of memory accesses in contemporary multi-processor systems, which identifies potential bottlenecks. However, in order to localize performance problems, it also has to be determined to which extend the application performance is limited by certain resources. Therefore, a methodology to derive metrics for the utilization of individual components in the memory hierarchy as well as waiting times caused by memory accesses is developed in the second step. The approach is based on hardware performance counters, which record the number of certain hardware events. The developed micro-benchmarks are used to selectively stress individual components, which can be used to identify the events that provide a reasonable assessment for the utilization of the respective component and the amount of time that is spent waiting for memory accesses to complete. Finally, the knowledge gained from this process is used to implement a visualization of memory related performance issues in existing performance analysis tools. The results of the micro-benchmarks reveal that the increasing number of cores per processor and the usage of multiple processors per node leads to complex systems with vastly different performance characteristics of memory accesses depending on the location of the accessed data. Furthermore, it can be observed that the aggregated throughput of shared resources in multi-core processors does not necessarily scale linearly with the number of cores that access them concurrently, which limits the scalability of parallel applications. It is shown that the proposed methodology for the identification of meaningful hardware performance counters yields useful metrics for the localization of memory related performance limitations.

Handling mobility in wireless sensor networks can pose formidable challenges in protocol design, especially, at the link layer. Since most of the proposed Medium Access Control (MAC) protocols do not accommodate mobility, a node has two options to deal with a deteriorating link: (a) to continue data transmission until the link breaks and then establishes a new link with a new relay node; or (b) to seamlessly transfer the communication to a better link parallel to the data transmission over the existing link. Different from option (a) where a node can only search for a new link after the original link disrupts, option (b) enables a node to perceive the change in the quality of a link in advance. The link quality prediction is implemented by an adaptive handover mechanism. Both approaches will inevitably introduce latency. This thesis aims to quantify and compare such latency. Specifically, it investigates the latency of packet transmission in a mobile wireless sensor network with and without the support of a handover mechanism. To start with, the thesis elaborates the effect of mobility on the performance of the existing MAC protocols, and the need to maintain an unbroken link during data transmission. To implement the handover, a target MAC protocol is required to be selected first. Since the Receiver-Initiated MAC protocol (RI-MAC) uses only short beacon and data packets during communication that substantially reduce overhearing, collision probability and data recovery cost, it is chosen as the carrier for the latency evaluation. Even though RI-MAC performs well in many aspects, it has several demerits. To address the monotonous increment in the backoff window size and to reduce the occurrence frequency of the dwell time, a burst data transmission pattern is adopted to optimize RI-MAC. With the optimization, the protocol reduces the long idle listening time that a node has to wait before data transmission, and thus, works well in a static scenario. However, due to the high probability of link disconnection, the burst data transmission does not perform well in case of mobility. For the sake of accommodating mobility, an adaptive handover mechanism is developed on top of the optimized RI-MAC. Once a node evaluates that the data packets cannot be completely transmitted before the link terminates, it will search for a new relay node while keeping communicating with the original collaborator. It is implemented by embedding a neighbor discovery request in a data packet that will be transmitted in a broadcast channel. Neighbors of the node will participate in the handover process as long as they are in an active state and their distance to the transmitter does not exceed a pre-defined threshold. As a proof-of-concept for the handover mechanism, a mathematical model is established. The transmission rate, the moving speed of human beings, the duty cycle and the network density are all taken into consideration. The analytical result shows that the communication latency decreases with an increment in the network density and the duty cycle when the handover mechanism is used, whereas the latency exhibits a reverse trend when the handover mechanism is not used. To validate the mathematics-based evaluation, the NS2 network simulator is employed. The simulation result is found to perform in accordance with the analytical result. It is asserted that the latency of packet transmission with the handover support is much less than that without the handover support. The communication latency can be saved by at least 0.28s when the handover mechanism is applied. This figure can even grow as the duty cycle and the network density increase. From this perspective, the handover mechanism is verified to improve the latency of packet transmission as far as mobility is concerned.

Das Internet of Things (IoT) ist aktuell ein junger Wachstumsmarkt, dessen Bedeutung für unsere Gesellschaft in naher Zukunft vielen Menschen erst noch wirklich bewusst werden wird. Die Subdomänen Smart-Home, Smart-Grid, Smart-Mobility, Industrie 4.0, Smart-Health und viele mehr sind wichtig für unsere zukünftige Wettbewerbsfähigkeit, die Herausforderungen zur Bewältigung des Klimawandels, unsere Gesundheit, aber auch für trivialere Dinge wie Komfort. Andererseits ergibt sich hierbei bereits dasselbe große Problem, das in einer ähnlichen Form schon bei klassischem Cloud-Computing bekannt ist: Vendor-Silos, die keinen hersteller- oder anbieterübergreifenden Austausch von Gerätedaten ermöglichen, verhindern eine schnelle Verbreitung dieser neuen Technologie. Diensteanbieter müssen ihre Produkte aufwendig für unzählige Technologien bereitstellen, was die Entwicklung von Diensten unnötig teuer macht und letztendlich das Dienstangebot insgesamt einschränkt. Cloud-Computing wird dabei in Zukunft eine wichtige Rolle spielen. Die Dissertation beschäftigt sich daher mit dem Problem IoT-Gerätedaten an IoT-Clouds plattformübergreifend und anbieterübergreifend nutzbar zu machen. Die Motivation und die adressierte Forschungslücke zeigen die Notwendigkeit der Beschäftigung mit dem Thema auf. Ausgehend davon, wird das Konzept einer dezentral organisierten IoT-Intercloud vorgeschlagen, welches in der Lage ist heterogene IoT-Clouds zu integrieren. Die Analyse des Standes der Technik zeigt, das IoT-Clouds genügend Eigenschaften teilen, um in Zukunft eine Adaption zu einer einheitlichen Schnittstelle für die IoT-Intercloud zu schaffen. Das Konzept umfasst zunächst die Komponentenarchitektur eines Intercloud-Brokers zur Etablierung einer IoT-Intercloud. Ausgehend davon wird in vertiefenden Teilkonzepten ein Discovery-Service zum Finden von Gerätedaten und einem Push-Stream-Provider, für die Zustellung von IoT-Event-Notifications in Echtzeit, behandelt. Eine Evaluation zeigt letztlich die praxistaugliche Realisierbarkeit, Skalierbarkeit und Performance der Konzeption und des implementierten Prototyps.

Systems for high performance computing are getting increasingly complex. On the one hand, the number of processors is increasing. On the other hand, the individual processors are getting more and more powerful. In recent years, the latter is to a large extent achieved by increasing the number of cores per processor. Unfortunately, scientific applications often fail to fully utilize the available computational performance. Therefore, performance analysis tools that help to localize and fix performance problems are indispensable. Large scale systems for high performance computing typically consist of multiple compute nodes that are connected via network. Performance analysis tools that analyze performance problems that arise from using multiple nodes are readily available. However, the increasing number of cores per processor that can be observed within the last decade represents a major change in the node architecture. Therefore, this work concentrates on the analysis of the node performance. The goal of this thesis is to improve the understanding of the achieved application performance on existing hardware. It can be observed that the scaling of parallel applications on multi-core processors differs significantly from the scaling on multiple processors. Therefore, the properties of shared resources in contemporary multi-core processors as well as remote accesses in multi-processor systems are investigated and their respective impact on the application performance is analyzed. As a first step, a comprehensive suite of highly optimized micro-benchmarks is developed. These benchmarks are able to determine the performance of memory accesses depending on the location and coherence state of the data. They are used to perform an in-depth analysis of the characteristics of memory accesses in contemporary multi-processor systems, which identifies potential bottlenecks. However, in order to localize performance problems, it also has to be determined to which extend the application performance is limited by certain resources. Therefore, a methodology to derive metrics for the utilization of individual components in the memory hierarchy as well as waiting times caused by memory accesses is developed in the second step. The approach is based on hardware performance counters, which record the number of certain hardware events. The developed micro-benchmarks are used to selectively stress individual components, which can be used to identify the events that provide a reasonable assessment for the utilization of the respective component and the amount of time that is spent waiting for memory accesses to complete. Finally, the knowledge gained from this process is used to implement a visualization of memory related performance issues in existing performance analysis tools. The results of the micro-benchmarks reveal that the increasing number of cores per processor and the usage of multiple processors per node leads to complex systems with vastly different performance characteristics of memory accesses depending on the location of the accessed data. Furthermore, it can be observed that the aggregated throughput of shared resources in multi-core processors does not necessarily scale linearly with the number of cores that access them concurrently, which limits the scalability of parallel applications. It is shown that the proposed methodology for the identification of meaningful hardware performance counters yields useful metrics for the localization of memory related performance limitations.

Handling mobility in wireless sensor networks can pose formidable challenges in protocol design, especially, at the link layer. Since most of the proposed Medium Access Control (MAC) protocols do not accommodate mobility, a node has two options to deal with a deteriorating link: (a) to continue data transmission until the link breaks and then establishes a new link with a new relay node; or (b) to seamlessly transfer the communication to a better link parallel to the data transmission over the existing link. Different from option (a) where a node can only search for a new link after the original link disrupts, option (b) enables a node to perceive the change in the quality of a link in advance. The link quality prediction is implemented by an adaptive handover mechanism. Both approaches will inevitably introduce latency. This thesis aims to quantify and compare such latency. Specifically, it investigates the latency of packet transmission in a mobile wireless sensor network with and without the support of a handover mechanism. To start with, the thesis elaborates the effect of mobility on the performance of the existing MAC protocols, and the need to maintain an unbroken link during data transmission. To implement the handover, a target MAC protocol is required to be selected first. Since the Receiver-Initiated MAC protocol (RI-MAC) uses only short beacon and data packets during communication that substantially reduce overhearing, collision probability and data recovery cost, it is chosen as the carrier for the latency evaluation. Even though RI-MAC performs well in many aspects, it has several demerits. To address the monotonous increment in the backoff window size and to reduce the occurrence frequency of the dwell time, a burst data transmission pattern is adopted to optimize RI-MAC. With the optimization, the protocol reduces the long idle listening time that a node has to wait before data transmission, and thus, works well in a static scenario. However, due to the high probability of link disconnection, the burst data transmission does not perform well in case of mobility. For the sake of accommodating mobility, an adaptive handover mechanism is developed on top of the optimized RI-MAC. Once a node evaluates that the data packets cannot be completely transmitted before the link terminates, it will search for a new relay node while keeping communicating with the original collaborator. It is implemented by embedding a neighbor discovery request in a data packet that will be transmitted in a broadcast channel. Neighbors of the node will participate in the handover process as long as they are in an active state and their distance to the transmitter does not exceed a pre-defined threshold. As a proof-of-concept for the handover mechanism, a mathematical model is established. The transmission rate, the moving speed of human beings, the duty cycle and the network density are all taken into consideration. The analytical result shows that the communication latency decreases with an increment in the network density and the duty cycle when the handover mechanism is used, whereas the latency exhibits a reverse trend when the handover mechanism is not used. To validate the mathematics-based evaluation, the NS2 network simulator is employed. The simulation result is found to perform in accordance with the analytical result. It is asserted that the latency of packet transmission with the handover support is much less than that without the handover support. The communication latency can be saved by at least 0.28s when the handover mechanism is applied. This figure can even grow as the duty cycle and the network density increase. From this perspective, the handover mechanism is verified to improve the latency of packet transmission as far as mobility is concerned.

Present-day and future network protocols that include and implement Forward Error Correction are configurable by internal parameters, typically incorporating expert knowledge to set up.We introduce a framework to systematically, objectively and efficiently determine parameters for Random Linear Network Codes (RLNC). Our approach uses an unbiased, consistent simulator in an optimization loop and utilizes a customizable, powerful and extendable parametric loss function. This allows to tailor existing protocols to various use cases, including ultra reliable, low latency communication (URLLC) codes. Successful configurations exploring the search space are under evolutionary pressure and written into a database for instant retrieval. We demonstrate three examples, Full Vector Coding, tail RLNC, and PACE with different focus for each.

Software systems in domains like Smart Cities, the Internet of Things or autonomous cars are coined by a high degree of distribution across several independent computing devices and the requirement to be able to adjust themselves to varying situations in their operational environment. Self-adaptive software systems are a natural choice to implement such context-dependent software systems. A multitude of approaches already implement self-adaptive systems and some consider even distribution aspects.Yet, none of the existing solutions supports the coordination of adaptation operations spanning multiple independent nodes, which is necessary to ensure a consistent adaptation even in presence of network errors or node failures. In this paper, we tackle this challenge to execute adaptations in distributed self-adaptive software systems in a coordinated manner. We present a protocol that enables the self-adaptive software system to execute correlated adaptations on multiple nodes in a transactional manner ensuring an atomic and consistent transition of the distributed system from its source to the desired target configuration. The protocol is validated to be free of deadlocks for any given adaptation at any point in time using a model-checking approach. The performance of our approach is investigated in experiments that emulate the protocol's execution on real devices for different sizes of distributed applications and adaptation scenarios.

Die parlamentarische Tradition in Sachsen geht bis in die vorkonstitutionelle Zeit zurück: Die Überlieferung spannt einen Bogen von den Versammlungen der Landstände des Spätmittelalters und der Frühen Neuzeit über die Verhandlungen des Zweikammerparlaments im Königreich Sachsen (1831/33 bis 1918) bis zu den Landtagen des Freistaates Sachsen in der Weimarer Republik (1919 bis 1933) sowie des Landes Sachsen in der Sowjetischen Besatzungszone und der DDR (1946 bis 1952). Ihre Fortsetzung fand diese parlamentarische Tradition mit der Wiederbegründung des Freistaates Sachsen im Jahr 1990. Zwischen 2007 und 2010 digitalisierte die SLUB Dresden mit Förderung des Sächsischen Landtages die gedruckten Protokolle der konstitutionellen Zeit ab 1831/33, seit Ende 2015 werden ergänzend die Landtagsakten bearbeitet. Der Vortrag auf einem Workshop zur Digitalisierung der Überlieferung der hessischen Landtage am 25. April 2018 im Institut für Personengeschichte Bensheim fasst die Erfahrungen des sächsischen Projektes zusammen und stellt das Onlineportal 'Historische Protokolle des Sächsischen Landtages' (https://landtagsprotokolle.sachsendigital.de) vor.

Präsenzbewertung im Kommunikationsraum der Telefonie ist ein zunehmend wichtiges Gebiet der Forschung. Ziel der Dissertation ist es, sich dem natürlichen Kommunikationsraum der verbalen und nonverbalen Kommunikation anzunähern. Im Kommunikationsraum Telefonie existieren noch keine Verfahren, die unterschiedliche Dringlichkeiten von Anrufen wiedergeben. Momentan ist die Signalart ausschließlich davon abhängig, wie der Angerufene sein Mobiltelefon eingestellt hat. In der Arbeit wurde ein Verfahren entwickelt, das unmittelbar vor der Signalisierung eines Anrufes die Signalart abhängig vom Kontext des Angerufenen individuell ermittelt und automatisch einstellt. Dabei werden Algorithmen aus dem Bereich der KI verwendet. Mit Hilfe von Lernalgorithmen werden die Ergebnisse der Präsenzbewertung kontinuierlich verbessert. Im Rahmen eines Experimentes konnten Präzisionen bei sechs verschiedenen Signalarten von bis zu 73 % erreicht werden. Dieses Verfahren ermöglicht eine deutliche Annäherung an den natürlichen Kommunikationsraum.

Das Internet of Things (IoT) ist aktuell ein junger Wachstumsmarkt, dessen Bedeutung für unsere Gesellschaft in naher Zukunft vielen Menschen erst noch wirklich bewusst werden wird. Die Subdomänen Smart-Home, Smart-Grid, Smart-Mobility, Industrie 4.0, Smart-Health und viele mehr sind wichtig für unsere zukünftige Wettbewerbsfähigkeit, die Herausforderungen zur Bewältigung des Klimawandels, unsere Gesundheit, aber auch für trivialere Dinge wie Komfort. Andererseits ergibt sich hierbei bereits dasselbe große Problem, das in einer ähnlichen Form schon bei klassischem Cloud-Computing bekannt ist: Vendor-Silos, die keinen hersteller- oder anbieterübergreifenden Austausch von Gerätedaten ermöglichen, verhindern eine schnelle Verbreitung dieser neuen Technologie. Diensteanbieter müssen ihre Produkte aufwendig für unzählige Technologien bereitstellen, was die Entwicklung von Diensten unnötig teuer macht und letztendlich das Dienstangebot insgesamt einschränkt. Cloud-Computing wird dabei in Zukunft eine wichtige Rolle spielen. Die Dissertation beschäftigt sich daher mit dem Problem IoT-Gerätedaten an IoT-Clouds plattformübergreifend und anbieterübergreifend nutzbar zu machen. Die Motivation und die adressierte Forschungslücke zeigen die Notwendigkeit der Beschäftigung mit dem Thema auf. Ausgehend davon, wird das Konzept einer dezentral organisierten IoT-Intercloud vorgeschlagen, welches in der Lage ist heterogene IoT-Clouds zu integrieren. Die Analyse des Standes der Technik zeigt, das IoT-Clouds genügend Eigenschaften teilen, um in Zukunft eine Adaption zu einer einheitlichen Schnittstelle für die IoT-Intercloud zu schaffen. Das Konzept umfasst zunächst die Komponentenarchitektur eines Intercloud-Brokers zur Etablierung einer IoT-Intercloud. Ausgehend davon wird in vertiefenden Teilkonzepten ein Discovery-Service zum Finden von Gerätedaten und einem Push-Stream-Provider, für die Zustellung von IoT-Event-Notifications in Echtzeit, behandelt. Eine Evaluation zeigt letztlich die praxistaugliche Realisierbarkeit, Skalierbarkeit und Performance der Konzeption und des implementierten Prototyps.:1 Einleitung 1.1 Problemstellung und Motivation 1.2 Ziele der Dissertation 1.2.1 Thesen 1.2.2 Forschungsfragen 1.3 Aufbau der Dissertation 2 Grundlagen zu Cloud-Computing im Internet of Things 2.1 Definition von Cloud-Computing 2.1.1 Generelle Eigenschaften 2.1.2 Architekturschichten 2.1.3 Einsatzformen 2.2 Internet of Things 2.2.1 Middleware im IoT 2.3 Architekturen verteilter Systeme zur Bereitstellung der IoT-Middleware 2.3.1 Geräte-zentrische IoT-Architektur 2.3.2 Gateway-zentrische IoT-Architektur 2.3.3 Cloud-zentrische IoT-Architektur 2.3.4 Zusammenfassung 2.4 Eigenschaften von verteilten Event-basierten Systemen 2.4.1 Interaktionsmodelle 2.4.2 Filtermodelle von Subscriptions 2.4.3 Verteiltes Notfication-Routing 2.5 Discovery im IoT 2.5.1 Grundlegende Begrifflichkeiten 2.5.2 Topologien von Discovery-Services 2.5.3 Funktionale Anforderungen für Discovery-Services im IoT 2.5.4 Ausgewählte Ansätze von Discovery-Services im IoT 3 Stand der Technik 3.1 Device-as-a-Service-Schnittstellen von IoT-Clouds 3.1.1 Gerätedatenmodell 3.1.2 Datenabruf mit Pull-Semantik 3.1.3 Datenabruf mit Push-Semantik 3.1.4 Steuerung von Gerätedaten 3.1.5 Datenzugriff durch Drittparteien 3.2 Analyse der DaaS-Schnittstellen verschiedener IoT-Clouds 3.2.1 Google Nest 3.2.2 Samsung Artik 3.2.3 AWS IoT 3.2.4 Microsoft Azure IoT Suite 3.2.5 Kiwigrid IoT-Plattform 3.2.6 Digi Device Cloud 3.2.7 DeviceHive 3.2.8 Eurotech Everyware Cloud 3.3 Zusammenfassung und Diskussion des Standes der Technik 4 Intercloud-Computing für das IoT 4.1 Intercloud-Computing nach Toosi 4.1.1 Ansätze zur Interoperabilität 4.1.2 Szenarien zur Cloud-übergreifenden Interoperabilität 4.1.3 Herausforderungen für Komponenten 4.2 Intercloud-Computing nach Grozev 4.2.1 Klassifikation der Architekturen 4.2.2 Klassifikation des Brokering-Mechanismus 4.2.3 Klassifikation verteilter Cloudanwendungen 4.3 Verwandte Arbeiten 4.3.1 Intercloud-Architekturen außerhalb der IoT-Domäne 4.3.2 Intercloud-Architekturen für das IoT 4.4 Analyse der verwandten Arbeiten 4.4.1 Systematik zur Bewertung 4.4.2 Bewertung und Abgrenzung 5 Anforderungsanalyse 5.1 Akteure in einer IoT-Intercloud 5.1.1 Menschliche Akteure 5.1.2 Systemakteure 5.2 Anwendungsfälle 5.2.1 Anwendungsfälle von IoT-Diensten 5.2.2 Anwendungsfälle von IoT-Clouds 5.2.3 Anwendungsfälle von IoT-Geräten 5.2.4 Anwendungsfälle von Intercloud-Brokern 5.3 Anforderungen 5.4 Ausschlusskriterien 6 Intercloud-Architektur für das IoT 6.1 Systemmodell einer IoT-Intercloud 6.1.1 IoT-Datenmodell für die Intercloud 6.1.2 Etablierung einer Vertrauensbeziehung zwischen zwei Clouds 6.2 Komponentenarchitektur des Intercloud-Brokers 6.2.1 Service-Connector, IC-DaaS-IF und Service-Protocol 6.2.2 Intercloud-Proxy, ICC-IF und Protokoll 6.2.3 Cloud-Adapter und IC-DaaS-Adapter-IF 6.3 Zusammenfassung 7 Verteilter Discovery-Service 7.1 Problembeschreibung 7.1.1 Topologie des Discovery-Service 7.2 Einfache Cloud-Discovery mit Broadcasting-Weiterleitung 7.2.1 Schnittstelle und Protokoll des einfachen Discovery-Service 7.2.2 Diskussion des einfachen Discovery-Service 7.3 Cloud-Discovery mit Geräteverzeichnis und Multicast-Weiterleitung 7.3.1 Geeignete Geräteinformationen für das Verzeichnis 7.3.2 Struktur und Schnittstelle des Verzeichnisses 7.3.3 Verzeichnissynchronisation und erweitertes Protokoll 7.4 Zusammenfassung beider Ansätze des Discovery-Service 8 Verteilter Push-Stream-Provider 8.1 Verteilter Push-Stream-Provider im Modell des Broker-Overlay-Netzwerks 8.2 Verteilter Push-Stream-Provider mit einfachem Routing-Modell 8.2.1 Systemmodell 8.2.2 Integration der Subkomponenten in die verteilte ICB-Architektur 8.3 Redundanz und Redundanzvermeidung des Push-Stream-Providers 8.3.1 Beschreibung des Redundanzproblems und des Lösungsansatzes 8.3.2 Lösungsansatz 8.4 Verteilter Push-Stream-Provider mit vereinigungsbasiertem Routing-Modell 8.4.1 Erkennen von ähnlichen Filtern 8.4.2 Konstruktion eines Vereinigungsfilters 8.4.3 Rekonstruktion der Datenströme 8.4.4 Komponente: Merge-Controller 8.4.5 Komponente: Stream-Processing-Engine 8.4.6 Integration in die bisherige Architektur 8.4.7 Diskussion des Ansatzes zur Redundanzvermeidung 8.5 Zusammenfassung zum Konzept des Push-Stream-Providers 9 Evaluation 9.1 Prototypische Implementierung der Konzeptarchitektur 9.1.1 Intercloud-Broker 9.1.2 IoT-Cloud und IoT-Geräte 9.1.3 IoT-Dienste 9.1.4 Grenzen des Prototyps und Fokus der experimentellen Evaluation 9.2 Aufbau der Evaluationsumgebung 9.3 Experimentelle Untersuchung der prototypischen Implementierung des Konzepts 9.3.1 Ermittlung einer Performance-Baseline 9.3.2 Experiment 1: Performance bei variabler Nachrichtengröße und Nachrichtenanzahl 9.3.3 Experiment 2: Performance bei multiplen Subscriptions 9.3.4 Experiment 3: Ermittlung des maximalen Durchsatzes und Skalierbarkeit des ICB 9.3.5 Experiment 4: Effizienzvergleich zwischen einfachem und vereinigungsbasiertem Routing 9.4 Zusammenfassung und Diskussion der Evaluation 10 Zusammenfassung 10.1 Beiträge der Dissertation 10.2 Ausblick A Abbildungen B Tabellen Inhaltsverzeichnis C Algorithmen D Listings Literaturverzeichnis