Analyse und Optimierung des Trade-offs zwischen QoE und Energieeffizienz in Datenzentren
Elektronische Halbleiter, Bauelemente und Schaltungen, Integrierte Systeme, Sensorik, Theoretische Elektrotechnik
Zusammenfassung der Projektergebnisse
During recent years, the number and size of data centers has constantly increased, as more and more services have become available “in the cloud”. As a result, energy-efficiency has also gained greater attention in operating a data center due to ecologic and economic consequences. However, saving energy usually leads to a degradation of the available computing resources which results to reduced Quality of Service (QoS). The diminished service quality can in turn lead to poor Quality of Experience (QoE) for the end-users. Although in literature several approaches address the optimization of data centers either to ensure energy-efficiency or to resolve certain QoS constraints, the joint optimization of data centers with respect to energy-efficiency and QoE beyond QoS is not researched. Further, related work on QoE mainly investigates multimedia applications only, e.g. the influence of transmission parameters in communication networks such as packet loss, delay or jitter on video QoE or voice-over-IP QoE. Being linked very closely to the subjective perception of the end user, QoE is supposed to enable a holistic understanding of the qualitative performance of networked communication systems and thus to complement the traditional, more technology-centric Quality of Service (QoS) perspective. This project focused on quantifying and adjusting the trade-off between QoE and energyefficiency in data centers for highly relevant use cases. An interesting use case for the interconnection between data centers, QoE, and energy-efficiency that is considered to have an increasing impact in the following years, are Virtual Desktop Infrastructures (VDIs). A VDI enables users to use very lightweight systems, e.g., so-called thin clients, whereby the actual operating system including all computations and software runs in a data center. The service migration to the cloud may lead to cost and energy savings due to economies of scale by better utilization of existing resources in the data center. However, if too many services are aggregated onto very few servers, the service requirements (e.g. in terms of processing time) cannot be fulfilled or result in significantly increased processing times. As a result, service quality and therefore QoE suffers. Additional challenges arise, as VDIs provide the user an entire operating system including various applications with different requirements. Common applications in this field are typical professional enterprise applications beyond multimedia including among others text processors, presentation software or spreadsheets software for which no QoE models exist in literature so far. Thus, in this project we first classified different application types according to their functionality and possible requirements on the client device, transmission network, and data center, for example, file storage services, online word processing tools, as well as enterprise systems, which are running inside VDIs. By classifying the file storage service Dropbox, for example, we found that beginners use Dropbox mainly for collaboration and file sharing while the synchronization users make a greater use of it for synchronization and backup. Second, detailed models for QoE in a VDI environment were derived based on subjective user studies e.g. via crowdsourcing. We found, among other things, that for Google Docs packet loss below 4% does not influence the duration of sub-processes, if there is no network delay. In contrast to this, network delay negatively influenced the performance and thus the QoE of Google Docs, even in the absence of packet loss. additionally, by evaluating an enterprise system, we investigated that also here technical monitoring parameters and subjective employee ratings are connected. These QoE model of concrete application over VDI quantifies the relation between the end-user QoE and different influence factors on the entire transmission chain including the client, network, and data center side. Using them, we were able to estimate the QoE of the given applications. Finally, these models were used for optimizing the QoE of file synchronization cloud services, video streaming services. Here, for example, we investigated models and trade-offs for virtualizing components of the mobile core network. These models can serve as a baseline reference to plan and dimension mobile network accordingly, not just based on expected user traffic as traditionally. Using our developed models of different application types, the optimal trade-off between QoE and energy-efficiency can be derived by adjusting control knobs of those mechanisms like scheduling of service requests, hot and cold standby of servers, or virtual machine (VM) placement approaches. To sum up, in this project we have taken a major step forward in analyzing and optimizing the trade-off between QoE and energy-efficiency in data centers. In parallel, the Hanoi University of Science and Technology (HUST) planned to define appropriate models for data center energy consumption based on existing studies. Assuming that all information on the capabilities of the users’ devices, the characteristics of the transmission network, and the capabilities of the data center architecture and the data center components would be available, this allows the analysis and adjustment of the trade-off between high QoE for both the end users and low power consumption in the data center side.
Projektbezogene Publikationen (Auswahl)
- “GTP-based Load Model and Virtualization Gain for a Mobile Network’s GGSN.” in: 5th International Conference on Communications and Electronics. Da Nang, Vietnam, July 2014
Florian Metzger, Christian Schwartz, and Tobias Hoßfeld
(Siehe online unter https://doi.org/10.1109/CCE.2014.6916704) - “Performance Model for Waiting Times in Cloud File Synchronization Services.” In: 26th International Teletraffic Congress (ITC). Karlskrona, Sweden, Sept. 2014
Christian Schwartz, Matthias Hirth, Tobias Hoßfeld, and Phuoc Tran-Gia
(Siehe online unter https://doi.org/10.1109/ITC.2014.6932939) - “Analyzing the Impact of Delay and Packet Loss on Google Docs.” In: 7th International Conference on Mobile Networks and Management. Santander, Spain, Sept. 2015
Lam Dinh-Xuan, Christian Schwartz, Matthias Hirth, Florian Wamser, and Huong Truong Thu
(Siehe online unter https://doi.org/10.1007/978-3-319-26925-2_16) - “Concept for Client-initiated Selection of Cloud Instances for Improving QoE of Distributed Cloud Services.” In: ACM SIGCOMM Workshop on QoE-based Analysis and Management of Data Communication Networks (Internet-QoE). Florianópolis, Brazil, Aug. 2016
Florian Wamser, Michael Seufert, Steffen Höfner, and Phuoc Tran-Gia
(Siehe online unter https://doi.org/10.1145/2940136.2940143) - “Modeling the YouTube Stack: from Packets to Quality of Experience.” In: Computer Networks 109.2 (Nov. 2016), pp. 211–224
Florian Wamser, Pedro Casas, Michael Seufert, Christian Moldovan, Phuoc Tran-Gia, and Tobias Hoßfeld
(Siehe online unter https://doi.org/10.1016/j.comnet.2016.03.020)
