Fakultätsübergreifende Veröffentlichungen
Permanent URI for this collection
Browse
Recent Submissions
Item Effiziente Kapazitätsplanung durch dynamische Erweiterung einer lokalen Ressourcenumgebung um Grid- und Cloud-Ressourcen: algorithmische und technische Betrachtungen(2014-06-04) Fölling, Alexander; Schwiegelshohn, Uwe; Kranzlmüller, DieterThe increasing complexity of scientific problems and industrial challenges leads to significant demands on computing infrastructures. Today, many scientific institutions and also most companies are still hosting their own computing infrastructures as commodity clusters or parallel machines to handle the upcoming load. Usually, these infrastructures are shared between many users with various applications. Caused by new problems and a fluctuating user community, changing load leads to phases with overloaded resources and undesired delays in job calculation on the first hand, and to idle phases with inefficent use of the available performance on the other hand. Instead of adapting the physical resources to the varying load, a resource operator is able to make use of algorithms and technologies of Grid- and Cloud- Computing to dynamically exchange load with delegation partners or a public cloud. This work investigates algorithmic approaches for the dynamic load exchange between autonomous resource environments in Computational Grids. The bilateral exchange between equitable partners is optimized using computational intelligence methods and subsequently is evaluated concerning improvements in service quality for the respective user communities. Usually, in Computational Grids the participating resource centers do not charge accepted workload. However, a cloud provider does charge unilateral load exchange. Thus, from the delegating centers point of view the charge represent an additional objective. This leads to a Multi-Objective Optimization Problem with the contradicting objectives of minimizing charge and maximizing service quality. In addition to common offline optimization algorithms, we introduce an adaptive reinforcement learning based policy to establish a beneficial relation of charge and gain for a considered resource centre. All presented algorithms are evaluated with real workload traces and examined regarding their robustness on foreign workload data or different local scheduling systems. The achieved results motivate the usage of dynamic load exchange algorithms for the improvement in service quality. Further, they serve as alternative to an extension of physical resources, which would be unfavorable from the economical and ecological perspective.Item Reflexion eines Mathematik-Vorkurses aus Teilnehmer- und Dozentenperspektive(2014-05-02) Klinger, MarcelDas vorliegende Schriftstück basiert auf den Erfahrungen im Rahmen eines Mathematik-Vorkurses zur Überbrückung der Diskrepanz zwischen schulischer und universitärer Mathematik, welcher vom 10. bis 28. März 2014 an der Fakultät für Informatik der Technischen Universität Dortmund stattfand. Der 15-tägige Brückenkurs bestand aus einem Vorlesungsteil sowie einer freien Lernumgebung mit anschließender Frontalbesprechung. Im Rahmen mehrerer anonymer Evaluationen wurden dabei neben personenbezogenen Grunddaten Einstellungen und Beliefs der Teilnehmer zur Mathematik sowie zum speziellen Vorkurs-Curriculum erhoben.