This research deals with optimal resource management in an overloaded broadband multimedia network. Optimisation is with respect to user satisfaction, where user satisfaction reflects both the quality of service experienced by connected users and the dissatisfaction of users blocked from access to the network. The research focuses on Asynchronous Transfer Mode (ATM) networks and the Internet, because these are the dominant emerging broadband networks which present some fundamental unsolved problems, related to the sharing of resources between mixed traffic types. ATM networks use conservative admission control, which protects network resources and ensures a high level of service for those admitted to the network, but results in low network efficiency because of low utilisation of resources due to blocking of many potential users. The Internet does not use admission control, with the result that performance degrades progressively as load increases. This causes frustration among users, and lowers the network efficiency due to high levels of congestion. We propose an optimisation model for each network (ATM networks and the Internet)which is intended to represent the distribution and consumption of key network resources by different traffic types. The model is aimed at maximising performance such that users admitted to the network are offered no less than some minimum acceptable level of quality of service (QoS). The solution is a set of traffic flow rates on each path which results in maximising an objective function value (revenue based on network operator interest or throughput based on customer interest) for a given network configuration with given user demand. As an example using the ATM network model, we illustrate the application of the model to an ATM network carrying both connection oriented and connectionless traffic. We explore the optimal response to a link failure which in turn causes node overload. As an example using the Internet model, we consider an overloaded network with link bottlenecks and an overloaded Web server, and explore the effect of transferring some server capacity to a mirror site and a proxy server. For real-time traffic control, the optimisation model is used to assign quotas for bandwidth or connections to selected paths. A control algorithm is implemented to provide maximum performance by admitting requests within the quotas which are obtained from the optimisation model. In an ATM network simulation, the algorithm is used to manage the virtual path (VP) pool in a network which suffers a link failure. A comparison is made between fixed virtual path management (FVPM) and dynamic virtual path management (DVPM), comparing the revenue achieved by each. This illustrates how DVPM adapts the VP pool in a robust fashion to achieve maximum revenue in the face of a link failure. However, the transient response suggests that benefit could be obtained using non-steady-state solutions. The model is extended by taking network state and traffic parameters into account to control changes in the VP pool to recognise limits to the rate at which traffic can be moved (through the natural birth-death processes). This scheme is called state dependent virtual path management (SDVPM). Performance evaluation of the new model shows that SDVPM achieves higher revenue than DVPM when the network suffers a link failure that requires a major change to the VP pool. In an Internet simulation, two algorithms are compared for control of access to a proxy server and a set of primary servers. An algorithm based on optimal flow solutions provides substantially better network performance than a localised heuristic algorithm. In each simulation case (ATM and Internet examples), the performance using a control system based on the steady state optimum flow model is close to the ideal optimal result.

Multimedia applications such as video streaming and Voice over IP are becoming common today with the tremendous growth of the Internet. General purpose operating systems thus are required to support these applications. These multimedia applications have some timing constraints that need to be satisfied for good quality. For example, video streaming applications require that each video frame be decoded in time to be displayed every 33.3 milliseconds. In order to satisfy these timing requirements, general purpose operating systems need to have fine-grained scheduling. Current general purpose operating systems unfortunately are designed to maximize throughput to serve traditional data-oriented applications and have coarse-grained scheduling and timers. Time Sensitive Linux (TSL), designed by Goel, et al., solves this problem with fine-grained timers and schedulers. The scheduler for TSL is implemented at a very low level. The controller that implements the algorithm for resource allocation is implemented at a higher level. This controller can easily be modified to implement new control algorithms. Successful implementation of resource allocation to satisfy timing constraints of multimedia applications requires two problems to be addressed. First, the resources required by the application to satisfy the timing constraints should not exceed the total available resources in the system. Second, the controller must adapt to changing needs of the applications and allocate enough resources to satisfy the timing constraints of each application over time. The first problem has been addressed elsewhere using intelligent data dropping with TSL. We focus on the second problem in this thesis. We design a proportion-period controller in this thesis for allocating CPU to multimedia video applications with timing constraints. The challenges for the controller design include the coarse granularity of the time-stamp markings of the video frames, the unpredictable decoding completion times of the frames, the large variations in the decoding times of the frames, and the limit of the control actuation to positive values. We set up the problem in a state space. We design a predictive estimating controller to allocate the proportion of the CPU to a thread when its long term error is small. When the decoding process is running behind by more than a certain threshold, we switch to a different controller to drive the error back to a small value. This controller is the solution to a dynamic optimization LQR tracking problem. / Graduation date: 2003

In order to leverage the use of various modalities such as audio-visual-touch in instilling learning behaviour, we present an intuitive approach of annotation based hapto-audio-visual interaction with the traditional digital learning materials such as eBooks. By integrating the traditional home entertainment system and respective media in the user's reading experience combined with haptic interfaces, we examine whether such augmentation of modalities influence the user's reading experience in terms of attention, entertainment and retention. The proposed Haptic E-Book (HE-Book) system leverages the haptic jacket, haptic arm band as well as haptic sofa interfaces to receive haptic emotive signals wirelessly in the form of patterned vibrations of the actuators and expresses the learning material by incorporating audio-video based augmentation in order to pave ways for intimate reading experience in the popular eBook platform. We have designed and developed desktop, mobile/tablet based HE-Book system as well as a semi-automated annotation authoring tool. Our system also supports multimedia based diverse quiz augmentations, which can help in learning tracking. We have conducted quantitative and qualitative tests using the developed prototype systems. We have adopted the indirect objective based performance analysis methodology, which is commonly used for multimedia based learning investigation. The user study shows that, there is a positive tendency of accepting multimodal interactions including haptics with traditional eBook reading experience. Though our limited number of laboratory tests reveal, that haptics can be an influencing media in eBook reading experience, but it requires large scale real life tests to provide a concluding remarks.

The objective of the following research is to investigate the problem of information management and conveyed experience on the World Wide Web (WWW) when multi-modal sensors and media are available. After studying related areas of work about the web and heterogeneous media, it became apparent that one of the main challenges of the area is the semantic unification of heterogeneous media. This thesis will introduce an event-based model to semantically unify media. An event is defined as something of significance that takes place at a given time and location. Using this definition and the corresponding model, a system will be designed to illustrate practical use cases for events.

The explosive growth of the Internet during the last few years, has been propelled by the TCP/IP protocol suite and the best effort packet forwarding service. However, quality of service (QoS) is far from being a reality especially for multimedia services like video streaming and video conferencing. In the case of wireless and mobile networks, the problem becomes even worse due to the physics of the medium, resulting into further deterioration of the system performance. Goal of this dissertation is the systematic development of comprehensive models that jointly characterize the performance of transport protocols and media delivery in heterogeneous wireless networks. At the core of our novel methodology, is the use of analytical models for driving the design of media transport algorithms, so that the delivery of conversational and non-interactive multimedia data is enhanced in terms of throughput, delay, and jitter. More speciffically, we develop analytical models that characterize the throughput and goodput of the transmission control protocol (TCP) and the transmission friendly rate control (TFRC) protocol, when CBR and VBR multimedia workloads are considered. Subsequently, we enhance the transport protocol models with new parameters that capture the playback buffer performance and the expected video distortion at the receiver. In this way a complete end-to-end model for media streaming is obtained. This model is used as a basis for a new algorithm for rate-distortion optimized mode selection in video streaming appli- cations. As a next step, we extend the developed models for the aforementioned protocols, so that heterogeneous wireless networks can be accommodated. Subsequently, new algorithms are proposed in order to enhance the developed media streaming algorithms when heterogeneous wireless networks are also included. Finally, the aforementioned models and algorithms are extended for the case of concurrent multipath media transport over several hybrid wired/wireless links.

This thesis presents novel techniques to index multiple media streams in a digi- tally captured presentation. These media streams are related by the common content in a presentation. We use relevance curves to represent these relationships. These relevance curves are generated by using a mix of text processing techniques and distance measures for sparse vocabularies. These techniques are used to automatically detect slide boundaries in a presentation. Accuracy of detecting these boundaries is evaluated as a function of word error rates.

The number and scope of data driven streaming applications is growing. Such streaming applications are promising targets for effectively utilizing multi-cores because of their inherent amenability to pipelined parallelism. While existing methods of orchestrating streaming programs on multi-cores have mostly been static, real-world applications show ample variations in execution time that may cause the achieved speedup and throughput to be sub-optimal. One of the principle challenges for moving towards dynamic pipeline balancing has been the lack of approaches that can predict upcoming dynamic variations in execution efficiently, well before they occur. In this thesis, we propose an automated dynamic execution behavior prediction approach based on compiler analysis that can be used to efficiently estimate the time to be spent in different pipeline stages for upcoming inputs. Our approach first uses dynamic taint analysis to automatically generate an input-based execution characterization of the streaming program, which identifies the key control points where variation in execution might occur with respect to the associated input elements. We then automatically generate a light-weight emulator from the program using this characterization that can predict the execution paths taken for new streaming inputs and provide execution time estimates and possible dynamic variations. The main challenge in devising such an approach is the essential trade-off between accuracy and overhead of dynamic analysis. We present experimental evidence that our technique can accurately and efficiently estimate dynamic execution behaviors for several benchmarks with a small error rate. We also showed that the error rate could be lowered with the trade-off of execution overhead by implementing a selective symbolic expression generation for each of the complex conditions of control-flow operations. Our experiments show that dynamic pipeline balancing using our predicted execution behavior can achieve considerably higher speedup and throughput along with more effective utilization of multi-cores than static balancing approaches.

Thesis (Ph. D.)--University of Missouri-Columbia, 2000. / Typescript. Vita. Includes bibliographical references (leaves 202-221). Also available on the Internet.

Thesis (Ph.D)--Electrical and Computer Engineering, Georgia Institute of Technology, 2010. / Committee Chair: Madisetti, Vijay; Committee Member: Jayant, Nikil; Committee Member: Lee, Chin-Hui; Committee Member: Ramachandran, Umakishore; Committee Member: Yalamanchili, Sudhakar. Part of the SMARTech Electronic Thesis and Dissertation Collection.

No description available.