Hans, Ronny (2018)
QoS-aware Cloud Infrastructure Provisioning in Heterogeneous Environments.
Technische Universität Darmstadt
Dissertation, Erstveröffentlichung

Kurzbeschreibung (Abstract)

Over the last decades Information Technology (IT) has become an enabler for nearly all businesses from industrial production to finance. The IT resources required for these business activities are usually provided by local and remote data centers. Although most resources are still hosted in companies’ proprietary data centers, cloud computing initiated a paradigm shift in IT service provisioning from owning to leasing resources and services. Today, over 50% of German companies use cloud services while shifting services into the cloud has become an emerging trend. Cloud computing, which is often referred to as the fifth utility in addition to water, electricity, gas, and telephony, provides commoditized computation resources that are available any time on demand in the required quantity. However, in contrast to other commodities, a single quality level is insufficient for IT service provisioning. Instead, the required quality for a provided IT service depends on the various functional and non-functional requirements. For example, highly interactive applications such as cloud gaming require a high quality level in terms of latency. Providers of cloud services have to face a highly competitive market. Cost advantages in cloud computing are primarily achieved by utilizing large centralized data centers at low-cost locations. However, this kind of resource provisioning impacts the quality of service of different types of services such as the aforementioned interactive multimedia services that possess strict quality of service constraints. Hence, infrastructure providers have to face a trade-off between cost reduction and adherence to the required Quality of Service (QoS) attributes. Apart from how services are provisioned, the way of consuming IT services also changed substantially over the last years. Mobile devices have begun to replace locally installed desktop computers at an accelerated pace. By utilizing these mobile devices, service providers are confronted with two major challenges: (i) a cellular network connection, which potentially causes a higher and more fluctuating latency and (ii) severely limited resources compared to local Personal Computers (PCs). These two aspects restrict the utilization of multimedia services, e. g., cloud gaming. To address these challenges, we present two novel approaches for (i) resource planning on a global level for multiple services with heterogeneous QoS characteristics and (ii) the augmentation of the centralized cloud infrastructure with locally installed resources to provide viable multimedia services to mobile devices. As the first major contribution, we introduce the Cloud Data Center Selection Problem (CDCSP). This problem describes the data center placement and resource selection on a global scale. We consider the role of a cloud provider, who aims to dimension resources in a cost-minimal fashion under the consideration of multiple services with different QoS attributes. Based on a mathematical optimization model, we propose the exact solution approach CDCSP-EXA.KOM. Due to the high complexity and the resulting computational effort to find the optimal solution, we propose and analyze four heuristic approaches to identify the most appropriate one for the given problem. As a first heuristic, we propose an approach based on linear program relaxation, CDCSP-REL.KOM. Furthermore, to take the specific structure of the problem into consideration, we develop the custom tailored CDCSP-PBST.KOM approach, which is based on a prioritized processing of demands and supplies. To further improve the results, we combine multiple heuristics to a Best-of-Breed approach, named CDCSP-BoB. KOM. Finally, as a metaheuristic improvement procedure, we propose the tabu search approach CDCSP-TS.KOM. To assess the practical applicability and performance of these optimization approaches, we analyze them in detail and compare their performance in a quantitatively. The second major contribution of this work addresses the augmentation of the centralized cloud infrastructure with local resources to provide services to mobile devices. Therefore, we formulate the Dynamic Cloudlet Placement and Selection Problem (DCPSP), as a multi-period resource planning problem, which includes local characteristics, such as space for hosting resources and available network bandwidth. We focus on a cloud provider who aims to augment the centralized infrastructure using local resources to improve the QoS guarantees for mobile used applications. We formalize the problem as a mathematical optimization model and derive the exact solution approach DCPSP-EXA.KOM. Due to the high complexity that is caused by an optimization over many time slots, we propose the heuristic optimization approach DCPSP-HEU.KOM. We assess the performance of these two approaches by the means of quantitative evaluation. In summary, the contributions of this thesis provide the means for a cost-efficient and QoS-aware resource selection in cloud infrastructures. We contribute the formalization of the problems and algorithms to support the efficient planning of future cloud infrastructures in environments with a multitude of heterogeneous services on a global scale. Furthermore, to enable mobile users to consume multimedia cloud services, we propose an optimization model and algorithms to augment a global centralized infrastructure by local resource units.

Frage zum Eintrag

Redaktionelle Details anzeigen