Universität Stuttgart

Permanent URI for this communityhttps://elib.uni-stuttgart.de/handle/11682/1

Browse

Filters

201624

Settings

Start date: 2016End date: 2016Institute: search.filters.UBSInstitut.Institut für Parallele und Verteilte SystemePublication Type: search.filters.UBSTyp.Abschlussarbeit (Master)

Search Results

Now showing 1 - 10 of 24
  • Thumbnail Image
    ItemOpen Access
    Cost optimization for data placement strategies in an analytical cloud service
    (2016) Saleem, Muhammad Usman
    Analyzing a large amount of business-relevant data in near-realtime in order to assist decision making became a crucial requirement for many businesses in the last years. Therefore, all major database system vendors offer solutions that assist customers in this requirement with systems that are specially tuned for accelerating analytical workloads. Before the decision is made to buy such a huge and expensive solution, customers are interested in getting a detailed workload analysis in order to estimate potential benefits. Therefore, a more agile solution is desirable having lower barriers to entry that allows customers to assess analytical solutions for their workloads and lets data scientists experiment with available data on test systems before rolling out valuable analytical reports on a production system. In such a scenario where separate systems are deployed for handling transactional workloads of daily customers business and conducting business analytics on either a cloud service or a dedicated accelerator appliance, data management and placement strategies are of high importance. Multiple approaches exist for keeping the data set in-sync and guaranteeing data coherence with unique characteristics regarding important metrics that impact query performance, such as the latency when data will be propagated, achievable throughputs for larger data volumes, or the amount of required CPU to detect and deploy data changes. So the important heuristics are analyzed and evolved in order to develop a general model for data placement and maintenance strategies. Based on this theoretical model, a prototype is also implemented that predicts these metrics.
  • Thumbnail Image
    ItemOpen Access
    Kooperative Vorhersage der minimalen Anwendungsausführungszeit
    (2016) Kuhn, Julian
    Code Offloading Frameworks verbessern durch Auslagern von Programmteilen - auch Offloadingkandidaten genannt - auf Server die Leistung oder den Energieverbrauch von Geräten mit limitierten Ressourcen. Offloadingkandidaten werden dann ausgelagert, wenn mit Inbetrachtnahme der Übertragung des Kandidaten eine Einsparung im Vergleich zur rein lokalen Ausführung vorliegt. Die Entscheidung, ob Offloading stattfindet, hängt stark von der Ausführungszeit des Kandidaten ab. Im Fall von Methoden kann die Ausführungszeit je nach aktueller Parameterkonfiguration stark variieren. Da es in vielen Fällen unpraktikabel ist, für jede Parameterkombination Aufzeichnungen durchzuführen, ist die Verwendung von einfachen, historienbasierten Modellen zur Bestimmung der Ausführungszeit ungeeignet. Eine möglichst genaue Angabe der Ausführungszeit wird aber benötigt, um die Offloadingentscheidung korrekt zu treffen. Ziel der Arbeit war, die Vorhersage von Ausführungszeiten mit Hilfe von Machine Learning Modellen anhand verschiedener Testanwendungen- und Szenarien im Kontext des Code Off-loadings zu untersuchen. Außerdem wurde ein kooperativer Systementwurf vorgestellt und implementiert, der zur Verwaltung von Datensätzen, Vorhersagemodellen und deren Erstellung, sowie zur Vorhersage von Ausführungszeiten verwendet werden kann. Der Entwurf erweitert dabei bestehende Offloadingframeworks. Es konnte festgestellt werden, dass sich Machine Learning Algorithmen zur Vorhersage und insbesondere zum Verbessern der Offloadingentscheidung eignen.
  • Thumbnail Image
    ItemOpen Access
    Automatic splitting in data-parallel complex event processing systems
    (2016) Sanwald, Tim
    Parallel Complex Event Processing (CEP) systems handle today’s heavy loaded event streams from smart homes, network traffic systems or stock trading systems by distributing the incoming event stream to several pattern detection systems. The correct splitting is currently done by CEP experts which ensure the consistent splitting without generating false-positive or false-negative complex events in comparison with centralized CEP systems. In this work an approach is developed which automatically generates a splitting model from the pattern definition which ensures the consistent distribution without generating false positives or false negatives. This approach enables a parallel CEP system to be configured and used the same way as a centralized CEP system. Further, a method which combines window based splitting and key based splitting is presented to reduce the network load and the CPU load on pattern detection operators. The functionality of the automatic splitting and the optimization is validated with common CEP scenarios based on generated and real world data to ensure a wide applicability of the approach.
  • Thumbnail Image
    ItemOpen Access
    Erklärung fehlender Ergebnisse bei der Verarbeitung hierarchischer Daten in Spark
    (2016) Mayer, Karsten
    Es existieren einige Algorithmen, die Entwicklern bei der Fehlersuche bei einer Datenbankanfrage helfen. Diese Arbeiten beantworten, wieso bestimmte Daten nicht in der Ergebnismenge für eine Anfrage vorhanden sind oder bestimmte nicht erwartete Daten in der Ergebnismenge erscheinen (Why-not-Frage). Für Anfragesprachen, die hierarchische Daten unterstützen, bestehen bisher aber nur wenige Arbeiten. In dieser Arbeit wird untersucht, welche Besonderheiten es für Why-not-Fragen bei hierarchischen Daten gibt. Dazu wird betrachtet, welche besonderen Fragestellungen dafür möglich sind und wie diese geeignet beantwortet werden können. Dabei wird auch ein konkreter Algorithmus für Python entworfen und implementiert. Anhand von diesem kann mit Hilfe eines Beispiels untersucht werden, ob der Algorithmus effizient und effektiv genug ist Why-not-Fragen zu beantworten.
  • Thumbnail Image
    ItemOpen Access
    Caching concept for mobile engineering apps
    (2016) Steffl, Michael
    Mobile apps in the engineering domain, have to deal with data coming from Product Data Management (PDM)-Systems. This data contains details about the products that are very large. Geometry data like 2D or 3D Computer Aided Design (CAD) representations are included. To get the data, the apps use wireless mediums like WiFi or mobile data networks (LTE, 3G, etc.). Transferring large size data over these mediums take a lot of time and can be aborted through intermittent connectivity. Also the energy consumption increases through the long-lasting transfers. In this master thesis a concept is created that overcomes these problems. A cache on the client is used that stores the relevant data for a fast access. As the disk space on mobile devices is limited, the data that is cached has to be chosen well. Only the data that is currently needed should be stored in the cache and provided to the app. To reduce the waiting times these data should be there before it is explicitly requested. To make this possible the concept of this thesis provides preemptive caching (hoarding). Thereby, the data is cached that will probably be needed next. To decide what data is needed, context is used. The information coming from the environment of the client is used, to derive situations. With the help of these situations the data is determined that gets cached. Besides this context-aware strategy, a traditional way of caching where all requested data gets cached is used in the concept. Furthermore, this thesis addresses the caching mechanism in its entirety. It determines a policy for the replacement of not needed data to free space. Also a strategy for invalidating obsolete data in the cache is determined. Finally, a prototypical implementation of the concept within an existing mobile engineering app is presented. With the help of this prototype the concept is evaluated.
  • Thumbnail Image
    ItemOpen Access
    Exploring classification algorithms and data feature selection for domain specific industrial text data
    (2016) Villanueva Zacarías, Alejandro Gabriel
    Unstructured text data represents a valuable source of information that nonetheless remains sub utilised due to the lack of efficient methods to manipulate it and extract insights from it. One example of such deficiencies is the lack of suitable classification solutions that address the particular nature of domain-specific industrial text data. In this thesis we explore the factors that impact the performance of classification algorithms, as well as the properties of domain-specific industrial text data, to propose a framework that guides the design of text classification solutions that can achieve an optimal trade-off between accuracy and processing time. Our research model investigates the effect that the availability of data features has on the observed performance of a classification algorithm. To explain this relationship, we build a series of prototypical Naïve Bayes algorithm configurations out of existing components and test them on two role datasets from a quality process of an automotive company. A key finding is that properly designed feature selection techniques can play a major role in achieving optimal performance both in terms of accuracy and processing time by providing the right amount of meaningful features. We test our results for statistical significance, proceed to suggest an optimal solution for our application scenario and conclude by describing the nature of the variable relationships contained in our research model.
  • Thumbnail Image
    ItemOpen Access
    In-network packet priority adaptation for networked control systems
    (2016) Zinkler, Stephan
    Sharing the network between Networked Control System (NCS) having strict demands with respect to latency and jitter and applications only requiring best-effort service leads to multiple problems. An important task to consider is how to prioritize individual types of traffic in such a way that the necessary guarantees for an NCS to be stable can still be given. While there are ways to prioritize the more important control traffic of an NCS over best-effort traffic sharing the same network, a more sophisticated approach has to be found in order to handle multiple NCS sharing the highest priority. In this thesis, in-network priority scheduling applications with a global view on the network are developed in order to schedule and prioritize individual NCS such that their stability can be guaranteed while sharing the network between multiple NCS. This thesis deals with in-network packet priority scheduling for Networked Control Systems. Using Data Plane Development Kit (DPDK) to achieve a Network Function Virtualization (NFV) based approach, a priority scheduling application is implemented in a middlebox to handle continuous priorities. This application could be instantiated and migrated within the network while simultaneously using Software Defined Networking (SDN) to route the traffic to the respective nodes. Additionally, this approach is extended using SDN and OpenFlow to adapt priorities in-network. Using the eight internal perport queues of a switch, discrete priorities are used to schedule, and additionally adapt, the priorities on the switch. This approach could give the opportunity for priority-based routing by using the SDN-controller for routing decisions and configuring the switches. The evaluation of this thesis is done by simulating NCSs and emulating the network containing the middlebox. For this, a simulation of an inverted pendulum is implemented for which the use of DPDK is compared to standard sockets. It can be shown that DPDK is able to perform better due to less delay and jitter. The scheduling application is evaluated by comparing it to a round-robin scheduling approach. The result suggests that the application is able to keep multiple NCS more stable than it’s round-robin counterpart. Furthermore, it is able to stabilize a more unstable system faster and more effectively. While the maximum sampling time for a system with a pendulum having an initial angle of 35° was found to be 50ms for the round-robin scheme, the middlebox is able to keep the system stable until 120ms. The application using OpenFlow is evaluated with respect to the time it takes to configure the switch as well as the overhead imposed by the configuration compared to the number of NCS within the network.
  • Thumbnail Image
    ItemOpen Access
    Definition eines Datenhaltungskonzepts für Industrie 4.0 Anwendungen
    (2016) Strljic, Matthias
    Diese Arbeit umfasst die Entwicklung eines Datenhaltungskonzeptes für Industrie 4.0 Anwendungen. Hierfür wurden Datenformate entlang der Industrie 4.0 Wertschöpfungsketten betrachtet und mit den Rahmenbedingungen des Referenzarchitekturmodells Industrie 4.0 ein Datenmodell erstellt. Dieses Modell wurde anschließend mithilfe von AutomationML in ein maschinenlesbares Format umgesetzt, welches Assetinformationen innerhalb eines einheitlichen Datenformates abbildet und somit die Wertschöpfung von formatübergreifenden Industrie 4.0 Anwendungen vereinfachen soll.
  • Thumbnail Image
    ItemOpen Access
    Active exploration and identification of kinematic devices
    (2016) Mohrmann, Jochen
    As an important part of solving the lockbox problem, this thesis deals with the problem of identifying kinematic devices based on data generated using an Active Learning strategy. We model the belief over different device types and parameters using a discrete multinomial distribution. We discretize directions as a Geodesic sphere. This allows an isotropic distribution without being biased towards certain directions. The belief update is based on experience using a Bayes Filter. This allows to localize the correct states, even if an action fails to generate movement. Our action selection strategy aims to minimize the number of actions necessary to identify devices by considering the expected future belief. We evaluate the effectiveness of different information measures and compare them with a random strategy within a simulation. Our experiments show that the use of the MaxCE strategy creates the best results. We were able to correctly identify prismatic, revolute, and fixed devices in 3D space.
  • Thumbnail Image
    ItemOpen Access
    Workload-enabled content-based routing in software-defined networks
    (2016) Sarmah, Himanshu
    Publish Subscribe Paradigm is a recent development in the area of loose coupling communication among distributed components. The communication can be facilitated with the help of content filtering within the network, here the content being the published messages. Using content filtering in the underlying network routers (also known as brokers in publisher/subscriber paradigm) enables to save valuable bandwidth by forwarding only those contents which are of interests to the subscribers. With the introduction of software defined networking, it is now possible to directly install content filtering rules in the Ternary Content Addressable Memory (TCAM) of the routers in the network. The content filtering rules are created by combining binary representation of the contents and IPv4/IPv6 addresses from a predefined range of addresses. However, the expressiveness of the filtering rules is limited by the small number of bits available in the IP addresses. With increasing amount of published/subscribed messages, it is also required to have proper filtering to avoid unnecessary traffic in the network. Therefore, initial work was proposed to create filtering rules based on the workload i.e. amount of published and subscribed messages, although these techniques are based on a simulated publisher/subscriber filtering environment. To explore the feasibility of the initially proposed techniques on an actual Software defined networking environment is the primary purpose of this thesis. This thesis discusses the proposed techniques with respect to a software defined networking environment and also implements the mapping of these techniques to work with a software defined publisher/subscriber network and evaluates the scalability and efficiency of the proposed techniques and the cost of doing so, in terms reconfiguration overhead.