Autonomic Quality of Experience Management of Multimedia Networks Steven Latré and Filip De Turck Ghent University - IBBT - Department of Information Technology, e-mail: steven.latre@intec.ugent.be Abstract—The proliferation of multimedia services over access networks (e.g., IPTV or network-based Personal Video Recording) has introduced important new revenue potential for network and service providers but has also complicated the management burden. As a result, today’s management of multimedia networks is often too static to cope with the increasing quality requirements of multimedia services. A key point in these quality requirements is the quality as perceived by the end users, denoted as the Quality of Experience (QoE). In the thesis, we have introduced an autonomic management layer that optimizes the QoE of multimedia networks. We have studied several QoE optimizing techniques with respect to traffic adaptation, admission control and video rate adaptation. All these QoE optimizing techniques exhibit autonomic behavior as they continuously monitor the network to optimize their configuration and consequently optimize the QoE. Furthermore, we have investigated the coordinated deployment of these QoE optimizing techniques by focusing on the exchange of context between entities in the distributed autonomic management layer. Through extensive evaluation using both simulation and emulation on a large-scale testbed, we have shown that the proposed QoE optimizing techniques can successfully optimize the QoE of multimedia services. This QoE optimization was characterized in terms of metrics such as the number of admitted sessions and video quality. I. M OTIVATION The management of multimedia networks is challenged by three major trends in today’s Internet: (1) a drastic increase in size of the current Internet and popularity of multimedia services, which stresses the importance of scalable management approaches, (2) a larger heterogeneity of the devices with the popularity of devices such as mobile phones, tablet PCs, etc. and (3) the higher quality requirements of multimedia services when compared to static services in terms of packet loss, throughput and connection duration. Throughout our work, we have focused on the management of access networks of which multimedia services such as IPTV and network-based Personal Video Recording are the most important services. Current management of multimedia access networks is still centered around a manual and reactive process where actions are taken by an operator when network anomalies occur. This manual management is not only timeconsuming and costly but also increasingly complex given the growing scale of the Internet and complex multimedia service’s quality requirements. When characterizing the quality requirements of multimedia services, the quality as perceived by the end users, commonly denoted as the Quality of Experience (QoE), is of the utmost importance. Previous research has focused mainly on 978-1-4673-0269-2/12/$31.00 c 2012 IEEE Higher QoE requirements Lower QoE requirements Fu D ll H vid eo L e ow ide e or so ion lut rv QoE Managed Network Traffic flow adaptation Video rate adaptation Admission control Fig. 1. The QoE in a multimedia network can be autonomically managed through flow adaptation, admission control and video rate adaptation. the design of accurate QoE metrics [1] and monitoring and reporting tools [2]. Our work is the first that takes this QoE information as an objective to steer the control algorithms of an autonomic management framework. We do not focus on the design of new QoE metrics but use QoE and QoS parameters already available in today’s networks to optimize the QoE. As shown in Figure 1, our work optimizes the QoE of multimedia services using techniques on three different axes: 1) Traffic flow adaptation modifies the network delivery of a traffic flow by changing the transport configuration or by accompanying the application data with additional (e.g., redundant) data. 2) Admission control does not modify the data of a flow but admits or blocks new connections to avoid congestion. 3) Video rate adaptation alters the application data itself by changing the actual video quality level that is transmitted as part of the multimedia service. These QoE optimizing techniques are not orthogonal and the configuration of one technique may influence the configuration of the other. Therefore, our work also focused on how context generated by one technique can be exchanged with others, to facilitate the coordination between these techniques. The remainder of this article is structured as follows. The concept of QoE is more thoroughly defined together with autonomic management approaches in Section II. Section III discusses the hypothesis and research questions we addressed in the thesis, while Section IV presents our contributions. Finally, Section V highlights some future research directions. II. BACKGROUND A. Quality of Experience Brooks et al. [3] define QoE as follows: ’QoE is a measure of user performance based on objective and subjective psychological measures of using a service or product’. They describe the link QoE has with objective QoS and provide an overview of the QoS parameters that contribute to the overall QoE. Specific to multimedia (i.e., video-based) services and products, the QoE is mainly impacted by the quality of the source video and its delivery. Although the QoE is inherently subjective, the authors of [3] discuss the need for objective and quantitative metrics that provide an estimation of the subjective QoE as this allows comparing between products and services. Several types of video quality metrics have been proposed in the past [1], which differ in whether or not (a part of) the original video sequence is required for the metric’s calculation. Full-reference metrics have the highest correlation with the human visual system and are therefore best suited for the off-line evaluation of the performance of video delivery. B. Multimedia networks The increasing popularity of QoE demanding services has resulted in the standardization of architectures such as Diffserv [4] and Intserv [5], which incorporate techniques for classifying traffic and providing them with a differentiated behavior based on the service type. These architectures can provide Quality of Service (QoS) through mechanisms such as admission control and resource reservation. However, these architectures provide a very coarse grained QoS management and are typically configured statically in which they overdimension the network considerably to provide the QoS. As video services have the lion’s share of today’s access networks, a lot of research is being carried out on the optimization of the video delivery by introducing more video oriented mechanisms into the access network. This has lead to a plethora of video optimizing techniques being proposed of which specific FEC codes [6], caching techniques [7] and retransmission mechanisms [8] are only a few examples. However, these video optimizing techniques are typically configured statically and make important assumptions on the traffic characteristics of the video they are optimizing. C. Autonomic management The static management of today’s networks introduces a severe limitation on which QoE and QoS guarantees can provided. Many authors have argued that a more flexible management is required to address the dynamic requirements of novel services such as multimedia services. Such a dynamic management is often translated into the research towards autonomic network management [9]. The term autonomic refers to the autonomic nervous system, where an equilibrium is maintained between several biological processes, without the need for conscious directions. An autonomic network is directed through high-level goals (i.e., policies), which results in an automatic and sub-conscious translation of actual configurations to reach that goal (e.g., the configuration of the network). Jennings et al. [10] provide an elaborate discussion on the challenges faced in the autonomic management of communication networks. They discuss the need for intelligent control algorithms that coordinate the different autonomic entities. This is achieved through control loops that dynamically configure the network but are deployed in a loosely coupled distributed environment, where multiple autonomic entities work together to jointly manage the network. Furthermore, they argue that semantic approaches such as ontologies are a vital part of any future autonomic network and are required because of the heterogeneity of today’s network devices and management domains on one hand and a strive to more formal mechanisms in reasoning on the other hand. III. H YPOTHESIS AND RESEARCH QUESTIONS The main objective of the thesis is to prove the validity of the following hypothesis: The QoE of multimedia services can be considerably improved by introducing an intelligent and distributed autonomic management layer, spanning multiple management domains, which has the high-level goal of optimizing the services’ QoE through the automated deployment and coordination of QoE optimizing techniques. To validate the above hypothesis, the following research questions were addressed: 1) How much can the QoE delivery of multimedia services be optimised by introducing more autonomic behaviour into current management solutions? In multimedia networks, a QoE guaranteed delivery is complicated by the occurrence of several network anomalies such as packet loss, throughput drops or fairness issues. The static configuration of today’s management techniques (e.g., congestion control and admission control) in contrast with the growing heterogeneity of today’s video services (e.g., multiple video quality levels, different burstiness) introduces severe limitations on the effectiveness of the management. A more dynamic and autonomic management is therefore needed. 2) Which QoE gain can be obtained by using application specific actions, that alter the services’ QoE, in combination with traditional network specific actions, that alter the network delivery? What is the corresponding trade-off between more customers and obtained average QoE score? Video services are one of the few services that can deliberately change the services’ QoE (e.g., to allow more users at a reduced quality) by modifying the application data that is sent over the network. Although, at the video encoding level, techniques exist to dynamically change the video quality level, there is a need for network Other root AEs Other root AEs Multimedia network's root AE AE AE AE Coordination of AEs AE AE AE QoE Optimizing Technique video rate adaptation QoE Optimizing Technique admission control QoE Optimizing Technique traffic adaptation Monitor: Network load Monitor: Network load Act: Downscale video Monitor: packet loss + connection's throughput Act: Change admission control configuration Act: Add redundancy + penalize misbehaving clients Fig. 2. Typical multimedia network and a conceptual overview of how the algorithms on different axes are deployed on the network’s devices. solutions that can decide when to scale to which video quality level so that a network or service provider can control the QoE levels he offers to the customers. 3) How can multiple management entities in a federation, spanning multiple management domains, interact with each other to form an end-to-end QoE optimising delivery framework? Although the best management decisions can be made if all knowledge is centralized, centralizing the management is not an option for scalability, overhead and privacy reasons. Instead, different management entities, spanning multiple management domains, need to cooperate with each other in a so-called federation. The entities in a federation need to interact with each other in order to form an end-to-end QoE optimization from the service originator up to the end customer. IV. C ONTRIBUTIONS OF THE DISSERTATION Figure 2 illustrates the hierarchical autonomic layer, consisting of multiple management entities or Autonomic Elements (AEs), that we have introduced on top of a typical access network. While AEs between administrative domains may have peer-relations (e.g., as part of a federation), we focus on the AE functionality in the multimedia network itself, which is hierarchical. Such a hierarchical approach has significant advantages in terms of scalability and the delegation of management decisions, which we have described in [11]. At the bottom of the hierarchy, several QoE optimizing techniques are deployed. These QoE optimizing techniques are components which are able to restore QoE in a well-defined scenario (e.g., congestion or lossy links). At higher layers, these QoE optimizing techniques need to be coordinated to ensure that the configuration of one technique does not negatively influence the performance of another. In our work, we have made several contributions to the field of autonomic QoE management. To be able to accurately evaluate the QoE optimizing techniques, we have proposed both a simulator (Section IV-A) and a management framework (Sec- tion IV-B) for large scale testbeds that allows characterizing the QoE of videos. Furthermore, we have proposed novel QoE optimizing techniques that perform traffic flow adaptation for both unreliable (Section IV-C1) and reliable (Section IV-C2) data transfers, admission control (Section IV-D) or video rate adaptation (Section IV-E). Finally, we have focused on how context (e.g., monitor data, reasoning decisions) can be exchanged between AEs (Section IV-F) through the design of a context dissemination framework. A. Design of a QoE enabled simulator As the goal of this work was to optimize the QoE of multimedia services, the evaluation methodology particularly focused on accurately characterizing the QoE. As a single QoS metric is not sufficient for grasping the subjective nature of QoE, we used a full-reference video quality metric called the Structural Similarity Index [12], throughout the evaluation of all QoE optimizing techniques. This allows us to have an accurate view on the improvement of each QoE optimizing technique on the video quality as perceived by the end user, which, for video services, is the most important part of QoE. In our work, we have designed an NS-2 [13] simulator that is able to incorporate these QoE oriented experiments by emulating the transmission of real video sequences over a simulated network [14], [15]. In this environment, the packets that are being transmitted over the simulated network do not contain real data but are dummy packets based on trace files of the video. These trace files can be generated using a dump of the generated traffic of the video streamer. On the receiving end, the simulator reconstructs the received video sequences so that the effect of simulated network anomalies can be visualized through the playback of the video. Furthermore, it allows calculating any video quality metric to objectively quantify the QoE. Through the proposed simulation environment, we were able to perform large-scale simulations: simulating several hours of video playback and transmission takes less than 2 minutes. B. Experiment automation for large scale testbeds Besides simulation, several of our contributions were also evaluated on a large-scale test-bed facility of more than 100 nodes. Testbed facilities provide a means to set up large scale network topologies but offer only a limited functionality in managing the deployment of the experiment itself. In order to address this, we have proposed a management framework for automating the setup of experiments in [16], [17]. The proposed management framework is generic but has been used extensively in our work for setting up multimedia network based experiments where QoE metrics are important. The framework focuses on the emulation of user behaviour by enabling the definition of behavioural profiles. These behavioural profiles define how a customer behaves over time, meaning which services he accesses, with what frequency, etc. As this behaviour heavily depends on the type of service that is accessed, a pluggable approach is taken where it is possible to define a service model for each type of service that is accessed. Furthermore, the framework allows simplifying the experiments both spatially and temporally. The spatial reduction consists of a virtualization approach where multiple virtual nodes are mapped onto one physical testbed node. For the temporal reduction, the experiment can be sped up by a speedup factor to reduce the time required to emulate: this allows us to emulate the behavior of one day in one hour or less. In [16], we have shown that the management framework allows to map at most 50 virtualized nodes on a physical node with a speedup factor of 50, without introducing a significant loss in accuracy. As such, the management framework supports experiment sizes of up to 5,000 nodes, which can be emulated 50 times faster than real time. C. Traffic flow adaptation: improving the QoE by modifying the network data 1) Unreliable data transfers, QoE optimization on the last mile: For unreliable data transfers, packet loss either due to congestion or a lossy link is the most important trigger for QoE disruptions. In our work, we focused on the QoE optimization on the last mile (i.e., the access network’s link just before the home network) of UDP-based video services by deploying a QoE optimizing technique on the access node. The QoE optimizing technique combines an adaptive FEC mechanism with a video quality switching algorithm based on monitoring input from the home network. The reasoner dynamically tunes the optimal redundancy level of the FEC mechanism but also selects the maximum possible video quality level, given the required redundancy, to avoid network congestion. To decide upon which action to take, the reasoner uses a monitoring layer called the Monitoring Plane [18], which intercepts monitoring reports containing perceived packet loss levels from the home network. For the design of the reasoner, we proposed three distinct algorithms: one based on an analytical modeling of the problem and two based on a neural network, which is either trained using a supervised or reinforcement learning technique [19]. We showed how the three reasoning algorithms have a similar performance: as an increasing amount of packet loss is introduced on the last mile, the redundancy level is increased accordingly to maintain a perfect QoE level. Additionally, the reasoner accurately identifies network congestion and adaptively lowers the video quality level when needed. Therefore, all three algorithms are able to achieve the best possible QoE level, given the network disruptions. When comparing the three reasoning algorithms, we showed how the neural network-based algorithms have the additional advantage of self-learning [14]. Especially for the reinforcement learning approach [19], we showed how the reasoner can actually be automatically generated through simulation and the definition of a reward function, if the problem domain can be modeled analytically. 2) Reliable data transfers, enforcing network fairness between misbehaving receivers: For reliable data transfers, we presented a QoE optimizing technique that focuses on the robustness of congestion avoidance algorithms against misbehaving TCP receivers. As congestion avoidance techniques rely on the integrity of the end hosts, they are prone to abuse from TCP subscribers. TCP subscribers that deliberately ignore congestion warnings (e.g., ECN messages) can obtain a higher throughput at the cost of reliable TCP subscribers as they gradually increase their rate when reliable TCP subscribers lower their rate as response to a congestion signal. We designed a cognitive mechanism, which can be deployed on an intermediate router and has the goal of detecting and penalizing such misbehaving TCP subscribers to restore network fairness. The algorithm uses a combination of clustering and outlier detection to divide the connections into two groups based on flow-based statistics, derived from monitoring the ECN packets of each flow. Misbehaving connections are penalized by assigning them to a different AQM category that drops the ECN signals: as such, misbehaving connections are forced to lower their throughput sooner than their responsive counter parts. We have shown that (1) misbehaving TCP receivers are able to establish a throughput which is 4 times higher than well-behaving TCP receivers, (2) our cognitive mechanism is able to detect these misbehaving receivers with an accuracy of more than 80% if the throughput differences are small and 100% when the throughput differences are substantial and (3) the differentiated AQM treatment allows successfully limiting the gain misbehaving connections achieve compared to well-behaving connections to 2% and less, thus restoring the network fairness. D. Design of an admission control system, optimized for protecting multimedia services Although traditional Admission Control (AC) solutions have already been widely deployed, the popularity of multimedia services has introduced some novel challenges of deploying an effective AC system in an IPTV environment. The standardized AC solutions such as those defined in the Diffserv [4] and Intserv [5] architecture are too static to cope with the high dynamics in the traffic characteristics of today’s videos. The consequence is that they take a conservative approach in dimensioning the amount of resources required. This dimensioning is based on the peak rate of the video, which is a worst case scenario and therefore leads to a severe underutilisation of the network. To address this, research is currently moving towards Measurement Based Admission Control (MBAC). In an MBAC system, the network load is distributively measured at various demarcation points in the network and these measurements are collected by specific nodes that perform the admission decision. Currently, the IETF is standardizing an MBAC system called Pre-Congestion Notification (PCN) [20], which measures the network load at each node and signals whether or not congestion is imminent on that node, by comparing it with a threshold rate R, through the marking of packets. We have investigated PCN’s performance in protecting multimedia services in [21]. We have shown that, although using PCN for video services results in an increase in admitted sessions, the configuration of the system is also very complex and prone to either underutilisation or either over-admittance if 1000 E. A policy-based video rate adaptation mechanism AC techniques are far more powerful if the traditional binary decision of blocking or admitting a session is extended to enable a graceful degradation of the QoE as the network load increases. Video services are one of the few services that can exhibit such a graceful degradation as they can intentionally and willingly adapt their QoE through video rate adaptation. At the video level, protocols such as Scalable Video Coding (SVC) [25] are able to encode the video into different quality levels and allow a smooth low-processing way of downscaling the video. At the network-level, solutions are required to ensure an optimized choice between these different qualities. We designed an algorithm that allows a network provider to control the QoE levels customers receive as a function of the network load. Compared to the state of the art (e.g., the range of HTTP Adaptive Streaming (HAS) techniques currently deployed such as Apple Live Streaming [26]) our algorithm is specifically intended for network providers of an IPTV environment: they can use the algorithm to define a policy on which QoE levels are offered to the clients and as such make a global optimization for their network. In contrast, when using HAS, the clients are responsible for deciding on which rate they want. As such, the operator does not have any control on the QoE the clients receive and what is optimal for the client might not be a global optimal for the network in terms of revenue. The algorithm is deployed on any PCNenabled node and uses the measurement information generated by the PCN system to locally adapt video quality levels that are forwarded on that node. To control the received QoE levels, the provider needs to define policies in the form of so-called utility functions. These utility functions define the share of a video quality as a function of the network load. What is typically defined is that the share of the lower video quality levels is gradually increased, while the share of the higher video quality levels is gradually decreased, as the network 1 915 Excellent QoE 900 0.9 700 651 0.8 600 500 0.7 400 0.6 300 SSIM Score (QoE estimation) 800 Number of admitted sessions the traffic characteristics of the videos change or are not known beforehand. To address this, and avoid the use of another conservative configuration, we have augmented the standard PCN mechanism by adding an autonomic layer on top that automates PCN’s configuration based on the monitoring of the underlying video traffic characteristics. The autonomic layer consists of (1) a novel load measurement algorithm, which is more robust against the burstiness of the video’s bitrate [21], (2) an adaptive rate algorithm, which autonomically configures the R threshold based on the variability of the traffic [22] and (3) an optional buffering component which allows decreasing the burstiness of the traffic at the expense of an increase in delay [23]. We have evaluated both the standard and augmented PCN mechanism and have shown that the introduction of the autonomic layer allows deploying PCN in an IPTV environment and obtaining a higher network utilization without needing to worry about the traffic characteristics of the video. Moreover, based on the performance evaluation, we have derived several guidelines which enable the configuration of a PCN system for an IPTV environment [24]. Moderate QoE Poor QoE 200 0.5 130 98 100 37 0 0.4 RACS PCN A-PCN Scale50SD Number of admitted sessions Average SSIM Score Maximum SSIM Score ScaleHQWeb Fig. 3. Influence of various admission control configurations on the network utilisation and QoE. load increases. Evaluation results of different combinations of admission control and video rate adaptation are shown in Figure 3. The investigated combinations are: traditional admission control mechanisms as used in the RACS layer of the TISPAN [27] architecture, the standardized PCN mechanism (with and without the autonomic layer discussed in the previous section) and the combination with two video rate adaptation configurations: one case where, at a network load of 100%, 50% of the videos are scaled to an SD resolution and 50% to a high quality web video (Scale50SD) and another case where all videos are scaled to a high quality web video (ScaleHQWeb). The impact on the number of admitted connections and QoE, estimated through the SSIM score, is shown for each configuration. As shown, traditional admission control is able to achieve a high SSIM score but the number of admitted connections is very low (i.e., 37). Introducing an MBAC approach such as PCN results in an increase in network utilisation of 350%, but decreases the SSIM score to a poor QoE as the traffic characteristics were not properly known beforehand leading to a misconfiguration. The autonomic layer (A-PCN) allows restoring this SSIM score again, with the cost of a small reduction in network utilisation. Combining this MBAC approach with the discussed video rate adaptation algorithm can considerably increase the network utilization with a limited QoE drop. Note that this drop still corresponds with a moderate video quality. The difference with the QoE drop caused by the traditional PCN configuration is threefold. First, a higher network utilisation can be achieved with a smaller QoE drop. Second, the QoE drop is due to the transmission of a lower video quality as opposed to data loss, and finally, as the video rate adaptation algorithm only downscales video if the network load is high, the maximum experienced SSIM score is still the highest possible. As such, the results show how, by setting a policy, the combination of admission control and video rate adaptation allows for a fine grained tuning of the QoE levels customers receive, while avoiding any QoE deterioration due to congestion. F. A context authoring process that automates the context exchange between federated nodes The proposed QoE optimizing techniques described above all introduce a higher level of adaptiveness and intelligence to optimize the delivery of video. However, the configuration of one technique may influence the configuration of the other. For example, as a FEC mechanism adds overhead there must be a link with an MBAC system to ensure that admitted connections are protected. There is thus a need for a distributed coordination layer that is able to orchestrate the configuration of these QoE optimizing techniques and is able to decide which techniques need to be deployed when and where. One of the first requirements of such a coordination layer is the exchange of context. To be able to effectively coordinate the different QoE optimizing techniques, a distributed coordination layer needs to have the right knowledge available to base its coordination decisions on. Similarly, also the QoE optimizing techniques require knowledge (e.g., the perceived packet loss of the client) to perform their management tasks. In our work, we have investigated the problem of context exchange in federated environments from a generic perspective and have applied it to both a multimedia network environment [28], [29] and a cloud environment [30]. Previous work (e.g., Macedo et al. [31]) has argued the need for a federated knowledge plane layer and presented the concept of query adaptation techniques. In our work, we build further upon this work by extending the query adaptation techniques to more context-aware semantic approaches, which allow a more greater flexibility in adapting the queries as they allow reasoning on the context and thus deduce the most appropriate context to exchange. In a typical publish subscribe system, consumers of context subscribe to the context by defining filter rules: only if the context matches the filter rules, the producers forward the context to the consumers. We have proposed an autonomic context exchange framework that automates the generation of these filter rules. In the proposed framework, the different entities define their contextual requirements in an ontology. These contextual requirements can be made dependent on the context: through semantic reasoning a new set of filter rules is constructed each time the context itself changes. We have shown that this context exchange process enables a highly dynamic environment where the context to transfer can easily be defined through policies. Furthermore, we showed that the filter rule generation takes less than 200 ms, which makes it feasible to deploy in an on-line scenario. G. Publications The thesis is publicly available for download from [32]. The results of this research have been published in several journals and magazines: the contributions related to the QoE optimization on the last mile have been published in Elsevier Computer Networks [14] and the contributions concerning admission control and video rate adaptation have been published in IEEE Communications Magazine [24] and Springer JNSM [33]. Our work has also been published in the proceedings of well-known conferences and workshops in the community such as IEEE CNSM [16], IEEE IM/NOMS [29], IEEE Globecom [21], MACE [19], [28], IEEE ACNM [22], [34] and others [15], [17], [35], [36]. Moreover, the thesis resulted in significant contributions to publications on (i) QoE metric design and monitoring [18], [37], [38], [39], [40], [41], (2) semantic management [30], [42], [43], (3) architectural design [11], [44], [45], (4) queueing and scheduling [23], [46] and (5) cloud management [47]. In summary, this Ph.D. resulted in 29 publications of which 7 in journals and magazines. H. Collaborations The results of our work have been performed in close collaboration with partners from both industry and academia. This work has been carried out in the context of 4 European projects: FP6 IP MUSE, CELTIC RUBENS, FP7 STREP ECODE and FP7 STREP OCEAN. Moreover, a few publications are the result of collaborations with Prof. John Strassner. V. F UTURE PERSPECTIVES The research presented in this work has introduced several important new research directions. First, while we investigated a first aspect of federated management being the exchange of context between entities in the federation, additional research towards the management of federations is needed to ensure a QoE guaranteed delivery across administrative domains. Several important challenges remain including the research towards coordination algorithms between federated partners and service matchmaking algorithms. Second, although multimedia services have seen an increased popularity over the last years, a recent topic that has experienced an important boost is the cloud computing paradigm. We believe the concepts studied in the thesis can also be applied to the management of cloud environments, although likely new problems such as the overload of an application server in the cloud can occur. Furthermore, a management layer for clouds will need to cooperate with a network management layer. While a cloud environment was not studied in this work, we believe an important research direction is the translation of the obtained results to the paradigm of clouds and the investigation of their performance on cloud-specific cases such as server consolidation and application placement. Third, as this work focused on the design of the lower layers in the autonomic management architecture, being the QoE optimizing techniques, there is an increased need for distributed control loops on the higher layers. Therefore, important research steps need to be taken to come to a reasoning process with an emphasis on the distributed deployment. For this, it will be necessary to strike a balance between introducing more learning capabilities (e.g., through machine learning) and at the same time providing sufficient insight in the details of the reasoning process (e.g., through a semantic description of the loop’s functionality). By combining these distributed control loops with the results presented in the thesis, end-to-end QoE guarantees can be made across management domains, which is the ultimate goal of an autonomic management layer. ACKNOWLEDGMENT Steven Latré’s Ph.D. work is funded by grant of the Fund for Scientific Research, Flanders (FWO-V). R EFERENCES [1] W. Lin and C.-C. J. Kuo, “Perceptual visual quality metrics: A survey,” Journal of Visual Communication and Image Representation, vol. In Press, Corrected Proof, pp. –, 2011. [Online]. Available: http://www.sciencedirect.com/science/article/B6WMK-52302SC2/2/d91c9d7300ab887497f5d7714b078649 [2] D. De Vera, P. Rodrı́guez-Bocca, and G. Rubino, “Qoe monitoring platform for video delivery networks,” in Proceedings of the 7th IEEE international conference on IP operations and management, ser. IPOM’07. Berlin, Heidelberg: Springer-Verlag, 2007, pp. 131–142. [Online]. Available: http://dl.acm.org/citation.cfm?id=1775321.1775337 [3] P. Brooks and B. Hestnes, “User measures of quality of experience: why being objective and quantitative is important,” Network, IEEE, vol. 24, no. 2, pp. 8 –13, 2010. [4] K. Nichols, V. Jacobson, and L. Zhang, “A Two-bit Differentiated Services Architecture for the Internet,” RFC 2638 (Informational), Jul. 1999. [Online]. Available: http://www.ietf.org/rfc/rfc2638.txt [5] R. Braden, D. Clark, and S. Shenker, “Integrated Services in the Internet Architecture: an Overview,” RFC 1633 (Informational), Internet Engineering Task Force, Jun. 1994. [Online]. Available: http://www.ietf.org/rfc/rfc1633.txt [6] T.-Y. Huang, P. Huang, K.-T. Chen, and P.-J. Wang, “Could skype be more satisfying? a qoe-centric study of the fec mechanism in an internetscale voip system,” Network, IEEE, vol. 24, no. 2, pp. 42 –48, 2010. [7] J. Famaey, T. Wauters, and F. De Turck, “On the merits of popularity prediction in multimedia content caching,” in Integrated Network Management (IM), 2011 IFIP/IEEE International Symposium on, may 2011, pp. 17 –24. [8] A. Nafaa, M. Gaucher, and L. Murphy, “Rs-vod: Reliable and scalable vod services provisioning system using p2p retransmissions,” in GLOBECOM 2010, 2010 IEEE Global Telecommunications Conference, dec. 2010, pp. 1 –6. [9] N. Agoulmine, Autonomic Network Management Principles: From Concepts to Applications. Elsevier, 2010, iSBN: 978-0-12-382190-4. [10] B. Jennings, S. van der Meer, S. Balasubramaniam, D. Botvich, M. Foghlu, W. Donnelly, and J. Strassner, “Towards autonomic management of communications networks,” Communications Magazine, IEEE, vol. 45, no. 10, pp. 112 –121, 2007. [11] J. Famaey, S. Latré, J. Strassner, and F. De Turck, “A hierarchical approach to autonomic network management,” in Network Operations and Management Symposium Workshops (NOMS Wksps), 2010 IEEE/IFIP, april 2010, pp. 225 –232. [12] Z. Wang et al., “Image Quality Assessment: From Error Visibility to Structural Similarity,” IEEE Transactions on Image Processing, vol. 13, no. 4, pp. 600–612, April 2004. [13] “Ns-2, The Network Simulator,” [online] http://www.isi.edu/nsnam/ns/. [14] S. Latré, P. Simoens, B. De Vleeschauwer, W. Van de Meerssche, F. De Turck, B. Dhoedt, P. Demeester, S. Van den Berghe, and E. Gilon de Lumley, “An autonomic architecture for optimizing QoE in multimedia access networks,” Computer Networks, vol. 53, pp. 1587–1602, July 2009. [Online]. Available: http://dl.acm.org/citation.cfm?id=1540673.1541273 [15] S. Latré, F. De Turck, B. Dhoedt, and P. Demeester, “Scalable simulation of qoe optimization for multimedia services over access networks,” in Proceedings of ICOMP2007 the 2007 International Conference on Internet Computing, 2007. [16] S. Latré, W. Van de Meerssche, S. Melis, D. Papadimitriou, F. De Turck, and P. Demeester, “Automated management of network experiments and user behaviour emulation on large scale testbed facilities,” in Network and Service Management (CNSM), 2010 International Conference on, oct. 2010, pp. 254 –257. [17] S. Latré, F. De Turck, D. Papadimitriou, and P. Demeester, “A management framework for automating network experiments and user behaviour emulation on large scale testbed facilities.” in ServiceWave’10, 2010, pp. 209–210. [18] B. De Vleeschauwer, P. Simoens, W. Van De Meerssche, S. Latré, F. De Turck, P. Dhoedt, B Demeester, S. Van Den Berghe, E. Gilon, and T. Van Caenegem, “Autonomic QoE Optimization on the Access Node,” in Broadband Europe, 2007. [19] S. Latré, P. Simoens, W. Van de Meerssche, B. De Vleeschauwer, F. De Turck, B. Dhoedt, P. Demeester, S. Van den Berghe, and E. Gilon De Lumley, “Automated generation of knowledge plane components for multimedia access networks,” in MACE ’08: Proceedings of the 3rd [20] [21] [22] [23] [24] [25] [26] [27] [28] [29] [30] [31] [32] [33] [34] [35] [36] IEEE international workshop on Modelling Autonomic Communications Environments. Berlin, Heidelberg: Springer-Verlag, 2008, pp. 50–61. [Online]. Available: http://dx.doi.org/10.1007/978-3-540-87355-6 5 P. Eardley, “Pre-Congestion Notification (PCN) Architecture,” RFC 5559 (Informational), Internet Engineering Task Force, Jun. 2009. [Online]. Available: http://www.ietf.org/rfc/rfc5559.txt S. Latré, B. De Vleeschauwer, W. Van de Meerssche, F. De Turck, P. Demeester, K. De Schepper, C. Hublet, W. Rogiest, S. Custers, and W. Van Leekwijck, “Design and configuration of PCN based admission control in multimedia aggregation networks,” in Global Telecommunications Conference, 2009. GLOBECOM 2009. IEEE, 30 2009-dec. 4 2009, pp. 1 –8. S. Latré, P. Simoens, B. De Vleeschauwer, W. Van de Meerssche, F. De Turck, B. Dhoedt, P. Demeester, S. Van Den Berghe, and E. Gilon de Lumley, “Design for a generic knowledge base for autonomic QoE optimization in multimedia access networks,” in Network Operations and Management Symposium Workshops, 2008. NOMS Workshops 2008. IEEE, april 2008, pp. 335 –342. K. Roobroeck, S. Latré, T. Wauters, and F. De Turck, “Optimized network utilisation through buffering in pcn enabled multimedia access networks,” in Integrated Network Management (IM), 2011 IFIP/IEEE International Symposium on, may 2011, pp. 1243 –1249. S. Latré, K. Roobroeck, T. Wauters, and F. De Turck, “Protecting video service quality in multimedia access networks through PCN,” IEEE Communications Maagazine, 2011, accepted for publication. H. Schwarz, D. Marpe, and T. Wieg, “Overview of the scalable video coding extension of the h.264/avc standard,” in IEEE Transactions on Circuits and Systems for Video Technology In Circuits and Systems for Video Technology, 2007, pp. 1103–1120. R. Pantos and W. May, “HTTP Live Streaming,” 2011. [Online]. Available: http://tools.ietf.org/html/draft-pantos-http-live-streaming-07 ETSI TS 182 019, “Resource and Admission Control Sub-system (RACS); Function Architecture,” 2009. S. Latré, S. Verstichel, B. De Vleeschauwer, F. De Turck, and P. Demeester, “On the design of an architecture for partitioned knowledge management in autonomic multimedia access and aggregation networks,” in Modelling Autonomic Communications Environments, ser. Lecture Notes in Computer Science, J. Strassner and Y. Ghamri-Doudane, Eds. Springer Berlin / Heidelberg, 2009, vol. 5844, pp. 105–110, 10.1007/978-3-642-05006-0 9. [Online]. Available: http://dx.doi.org/10.1007/978-3-642-05006-0 9 S. Latré, S. van der Meer, F. De Turck, J. Strassner, and J. Hong, “Ontological generation of filter rules for context exchange in autonomic multimedia networks,” in Network Operations and Management Symposium (NOMS), 2010 IEEE, april 2010, pp. 575 –582. J. Famaey, S. Latré, J. Strassner, and F. De Turck, “A hierarchical context dissemination framework for managing federated clouds,” Journal of Communications and Networks, 2011. D. Macedo, A. dos Santos, J. Nogueira, and G. Pujolle, “A knowledge plane for autonomic context-aware wireless mobile ad hoc networks,” in Management of Converged Multimedia Networks and Services, ser. Lecture Notes in Computer Science, G. Pavlou, T. Ahmed, and T. Dagiuklas, Eds. Springer Berlin / Heidelberg, 2008, vol. 5274, pp. 1–13. S. Latré, Autonomic QoE Management of Multimedia Networks, on-line available at http://users.ugent.be/˜slatre/slatre PhDThesis.pdf , ISBN: 978-90-8578-429-6. S. Latré, B. De Vleeschauwer, W. Van de Meerssche, K. De Schepper, C. Hublet, W. Van Leekwijck, and F. De Turck, “PCN based admission control for autonomic video quality differentiation: Design and evaluation,” Journal of Network and Systems Management, vol. 19, pp. 32–57, 2011, 10.1007/s10922-010-9183-8. [Online]. Available: http://dx.doi.org/10.1007/s10922-010-9183-8 S. Latré, B. De Vleeschauwer, W. Van de Meerssche, S. Perrault, F. De Turck, P. Demeester, K. De Schepper, C. Hublet, W. Rogiest, S. Custers, and W. Van Leekwijck, “An autonomic PCN based admission control mechanism for video services in access networks,” in Integrated Network Management-Workshops, 2009. IM ’09. IFIP/IEEE International Symposium on, june 2009, pp. 161 –168. S. Latré, N. Staelens, F. De Turck, B. Dhoedt, and P. Demeester, “Improving the quality of multimedia services to wireless users through AHAFEC deployment.” in EuroFGI Workshop on IP QoS and Traffic Control, 2007. S. Latré, N. Staelens, P. Simoens, B. De Vleeschauwer, W. Van de Meerssche, F. De Turck, B. Dhoedt, P. Demeester, S. V. den Berghe, [37] [38] [39] [40] [41] R. Huysegems, and N. Verzijp, “On-line estimation of the QoE of progressive download based services in multimedia access networks.” in International Conference on Internet Computing’08, 2008, pp. 181– 187. P. Simoens, S. Latré, B. De Vleeschauwer, W. Van de Meerssche, F. De Turck, B. Dhoedt, P. Demeester, S. Van Den Berghe, and E. Gilon, “Design of an autonomic QoE reasoner for improving access network performance,” Autonomic and Autonomous Systems, International Conference on, vol. 0, pp. 233–240, 2008. S. Verstichel, S. Latré, S. Melis, F. De Turck, B. Dhoedt, and P. Demeester, “A perfsonar-compliant looking glass for the geant2 multidomain monitoring service,” in Proceedings of the International Conference on Parallel and Distributed Processing Techniques and Applications, PDPTA 2007, Las Vegas, Nevada, USA, June 25-28, 2007, Volume 1, H. R. Arabnia, Ed. CSREA Press, 2007, pp. 372–378. P. Leroux, S. Latré, F. De Turck, and P. Demeester, “Experimental characterisation of the impact of ip-based distribution network QoS on the QoE of DVB-H mobile broadcast video services,” in Integrated Network Management-Workshops, 2009. IM ’09. IFIP/IEEE International Symposium on, june 2009, pp. 263 –268. P. Leroux, S. Latré, N. Staelens, P. Demeester, and F. De Turck, “Mobile TV services through IP datacast over DVB-H: Dependability of the quality of experience on the IP-based distribution network quality of service,” Journal of Network and Computer Applications, vol. 34, no. 5, pp. 1474 – 1488, 2011, dependable Multimedia Communications: Systems, Services, and Applications. [Online]. Available: http://www.sciencedirect.com/science/article/pii/S1084804510001736 C. Develder, P. Lambert, W. Lancker, S. Moens, R. Van de Walle, J. Nelis, D. Verslype, S. Latré, N. Staelens, and N. Vercammen, “Delivering scalable video with QoS to the [42] [43] [44] [45] [46] [47] home,” Telecommunication Systems, 2010. [Online]. Available: http://www.springerlink.com/index/10.1007/s11235-010-9358-3 J. Famaey, S. Latré, J. Strassner, and F. De Turck, “An ontologydriven semantic bus for autonomic communication elements,” in Modelling Autonomic Communication Environments, ser. Lecture Notes in Computer Science, R. Brennan, J. Fleck, and S. van der Meer, Eds. Springer Berlin / Heidelberg, 2010, vol. 6473, pp. 37–50, 10.1007/9783-642-16836-9 4. [Online]. Available: http://dx.doi.org/10.1007/978-3642-16836-9 4 ——, “Semantic context dissemination and service matchmaking in future network management,” International Journal of Network Management, 2011. [Online]. Available: http://dx.doi.org/10.1002/nem.805 S. Verstichel, A. Baart, G. Lievens, S. Latré, F. De Turck, and F. Vermeulen, “Train management platform for advanced maintenance of passenger information systems,” in Telecommunications, 2007. ITST ’07. 7th International Conference on ITS, Jun. 2007, pp. 1–6. [Online]. Available: http://dx.doi.org/10.1109/ITST.2007.4295921 S. Van den Berghe, P. Nooren, S. Latré, B. Crabtree, M. Kind, E. Viruete, and C. Pons, “QoE-driven broadband access,” in Proceedings of the 2008 NEM Summit Conference,, 2008. J. Famaey, W. Van de Meerssche, S. Latré and, S. Melis, T. Wauters, F. De Turck, K. De Schepper, B. De Vleeschauwer, and R. Huysegems, “Towards intelligent scheduling of multimedia content in future access networks,” in Network Operations and Management Symposium (NOMS), 2010 IEEE, april 2010, pp. 821 –824. H. Moens, J. Famaey, S. Latré, B. Dhoedt, and F. De Turck, “Design and evaluation of a hierarchical application placement algorithm in large scale clouds,” in Integrated Network Management (IM), 2011 IFIP/IEEE International Symposium on, may 2011, pp. 137 –144.