[Back]

J. Matt:
"Dynamic Optimization of Data Object Placement in the Cloud";
Supervisor: S. Schulte, P. Waibel; Institute of Information Systems, Distributed Systems Group, 2017; final examination: 2017-05-31.

The use of cloud-based storages to store data has become a popular alternative to traditional local storage systems. Users of cloud-based storages can benefit from a lot of advantages, such as higher data availability, extended durability and lower IT
administration cost. However, there also exist drawbacks in using cloud-based storage systems. Among the biggest drawbacks are the problem of vendor lock-in and possible unavailability of the data. To overcome these problems, in this thesis we formulate a system model that makes use of multiple cloud storages to store data. The usage of this system model allows the redundant storage of data and aims at finding the cheapest possible storage solution.
In this thesis we formulate a global optimization problem that takes into account historical data access information and ensures predefined Quality of Service requirements to find a cost-efficient storage solution. Furthermore, we present a highly scalable heuristic
approach that can be used with big amounts of data. The heuristic approach aims at scalability while still providing a cost-efficient storage solution that comes close to the optimal solution provided by the global optimization. As an extension to the heuristic approach we also describe a latency consideration approach. This approach incorporates latencies between the middleware and the used cloud storages into the optimization in order to find a storage solution that offers the lowest possible latencies. Therefore, customers can access their data faster which leads to a better end-user experience.
We extensively evaluate all three optimization approaches and thereby show the benefits of our designed approaches. The evaluations are presented by comparing our optimization approaches to a baseline that follows a state-of-the-art approach. We prove the correct functionality of the approaches by a detailed analysis of the results and show that we save more than 35% of total cost in comparison with the baseline.