We present different topics of our research every Thursday at 12:00 noon in room 103, Obermarkt 17. For news about the EAD-Lunch talks and seminars please feel free to subscribe to EAD-Public@googlegroups.com. (Register here: https://groups.google.com/forum/?hl=de&fromgroups#!forum/ead-public)
Today most cloud providers like Google, Amazon, Digital Ocean or Rackspace, charge the customers for the used number of instance per hour or even per minute. Additionally, one can use nearly an infinite number of instances with each cloud provider. This makes it possible to scale the infrastructure several times a day in order to provide the best user experience at the lowest infrastructure costs. In this talk we present the development and implementation of concepts for flexible IT-Infrastructure management in the cloud.
This research work presents a general methodology for NP-hard scheduling problems. Due to the NP-hard nature of most of these problems, they have predominantly been solved utilizing the metaheuristic algorithms, consisting of Genetic Algorithm, Simulated Annealing, Ant Colony Optimization, etc.. Apart from these methodologies, these problems can also be tackled with Integer Programming (IP). However, due to the exponentially increasing number of decision variables, IP solvers fail to solve large sized problem instances on conventional computing devices. This work is based on splitting the 0-1 Integer Programming in two parts to basically reduce the size of the search space. This split leads to a linear program and a set of decision variables. Since Linear Programming (LP) is polynomially solvable, they can be integrated with the metaheuristic algorithms to obtain optimal/near-optimal solutions. However, using LP solvers with an iterative metaheuristic algorithm is time consuming as these solvers are not fast enough. Hence, an effective utilization of this approach requires the development of some fast specialized polynomial algorithms for the resulting LP. Developing these exact polynomial algorithms is not straight forward except for trivial cases, and requires theoretical analyses of the specific linear program at hand. In this work, we utilize this approach over several NP-hard scheduling problems mainly in the field of transportation and manufacturing. We develop novel specialized algorithms for the resulting LPs to exploit them in conjunction with the metaheuristic algorithms to provide optimal/near-optimal solutions. Another benefit of this approach is its inherent parallel structure which is demonstrated later in this work with the help of Graphics Processing Unit (GPU) computing. Moreover, we also discuss how this generalized approach can be extended to other combinatorial optimization problems, apart from scheduling.
E-Commerce is playing an increasingly important role for both, operators of shopping platforms and customers. The forecast revenue for the German E-Commerce market amounts to EUR 46.7 million, which equals to more than 10% of the total retail sector. 30% of the online purchases in Germany are done via mobile devices. Despite an increasing public awareness of the issue of data protection, nowadays only in the rarest cases customers are allowed to decide how and whether their personal information and buying behavior is stored and processed. This threat to the right to informational self-determination is further exacerbated by the fact that the increasing use of smartphones, tablets and other portable devices leads to an ever more detailed profiling and thus in a deeper engagement in the privacy of each individual. Based on this situation, the SIOC project´s vision is to improve self-data protection in E-Commerce. SIOC enables customers to perform the online-shopping process transparently, as anonymous as possible, following the principle of data thrift/minimization (collection limitation principle). At the same time online platform providers still have the possibility to submit personalized offers and recommendations based on anonymized, voluntary aggregated customer profiles. Moreover, in case of non-paying customers, anonymization can be revoked to clearly identify customers. Thus, SIOC objective is the design of an anonymous approach to online shopping in accordance to Stakeholders requirements and business models to achieve the best compromise between these conflicting interests while implementing data protection by design and default as essential principles of EU data protection rules.
As the number of web documents as well as the inherent data and information is growing at a rapid pace, the interest in extracting and utilizing this data is on the rise as well. The aspects that are unlocked by Web Data Extraction to its users are as endless as the extensiveness of topics and fields of work that are inherent within the Web. The major obstacle is to utilize the available data, contents and processes. Especially for the first two, Web Data Extraction tries to offer tools and solutions. Several research papers have already showed developments and approaches to solve Web Data Extraction tasks. However, to meet the requirements and complexities of current web data, we require better technologies. Additionally, when looking from an users perspective, there is still a huge gap between research results and real applicability of potential techniques. Available solutions including research results, commercial products or open source solutions lack certain capabilities or as suffer from severe usability issues. In this talk we review the state-of-the-art in this research field and survey over freely available solutions and their applicability to everyday extraction tasks.
Building a general model for anything can be quite challenging. It depends on the level of generalization which further depends on the field of application for this model and the ultimate goal that needs to be achieved with it. The talk will cover a brief recap of the previous research in cloud resource management to motivate the necessity of a general resource and application model for demand and runtime prediction in the cloud. The introduction is followed by a discussion about challenges for developing such a model eventually leading into a brainstorming session for this topic. Finally, means to overcome these obstacles will be identified.