Year

2006

Degree Name

Doctor of Philosophy

Department

School of Electrical, Computer and Telecommunications Engineering - Faculty of Engineering

Abstract

The rapid increases in network bandwidth and processing power have led to tremendous growth in Internet applications and changed the nature of delivering content. From earlier web application, which is only about retrieval of pre-computed static content, current content delivery architectures provide dynamic content and enable personalization of content. Recent interactive entertainment applications, such as multiplayer online games, require content to be created and distributed in real-time. In addition, the rapid increase in processor speed has led to various proposals to put real-time computation within or at the edge of the network that allow application specific processing on packet flow such as multimedia transcoding and content adaptation. In short, these emerging applications have a common characteristics that the contents of application flows are processed in real-time before being delivered to end users. We refer to these as applications that require real-time content creation.

This thesis aims to develop models for real-time content creation and distribution. We examine both network architectures for delivering content as well as server processing resource management for content creation. We concentrate on a case study when content creation is from a dynamic set of dispersed sources. In particular, we examine the provision of an immersive voice communication service to massively multiplayer online games, which requires real-time creation of audio scenes from dynamic sets of participants. We present various delivery architectures for this service, evaluate the performance of these architectures and provide recommendations based on the evaluation. In addition, this thesis designs a server resource management architecture for sharing processing resource among real-time content creation applications.

Our study begins with reviewing the evolution of content distribution over the Internet, ranging from simple caching proxies to content distribution networks and personalization of content. We then discuss current and future developments of content distribution which require real-time creation and distribution of content from dynamic sets of dispersed sources. These include state information processing and communication information processing in distributed virtual environments. While all networked games require state information processing, communication information processing has been recently seen as a key to enhance the reality and attractiveness of the virtual environment. In particular, this thesis reviews technologies and approaches for providing an immersive voice communication service to distributed virtual environments. Several delivery architectures are introduced for providing this service, namely peer-to-peer, central server, distributed locale server architecture, and distributed proxy architecture. Furthermore, we present a realistic simulation model that captures player distribution in the Internet and avatar distribution in the game virtual world and specify two key performance evaluation parameters: interactive delay and network bandwidth usage.

In the central server architecture, two optimization objectives are proposed for choosing an optimal central server from a set of potential servers. We also propose a dynamic relocation of a central server in response to changes in player distribution due to time zone differences. It is shown that relocation of the central server in response to these changes can significantly reduce the interactive delay by up to 40% and the network bandwidth usage by up to 50%. In addition, the optimal central server can significantly reduce the interactive delay compared to a randomly located central server. In the distributed locale server architecture, the game virtual world is partitioned into smaller areas called locales and each locale is assigned to a server. We propose two server assignment algorithms for optimizing the latency performance of this architecture. The first algorithm is based on an Integer Linear Programming (ILP) model which provides an exact solution to the problem but is subject to high computation complexity. We then produce a new multi-layer graph representation of the problem and devise a greedy heuristic based on this graph. It is shown that the greedy heuristics has low run time complexity and provides solutions close to the optimal (within 5% of the optimal in all cases). In addition, increasing the number of servers reduces the latency of the distributed locale server architecture significantly compared to the optimal central server when there is a physical/virtual world correlation. Specifically, with a reasonable number of servers, the distributed locale server architecture can reduce the delay of the central server by 20% to 60%.

In the distributed proxy architecture, players are assigned to a close proxy and each proxy manages the audio mixing operation on behalf of players and forwards audio streams from players to other interested proxies. This thesis develops an ILP model for an optimal proxy assignment and adapts the multi-layer graph approach used earlier to devise a greedy heuristics for solving the proxy assignment problem efficiently. While the ILP model is unscalable, the greedy heuristics is highly scalable and suitable for practical implementation. This thesis also investigates the efficiency of network multicast in different player and avatar distribution scenarios. The effect of varying the number of proxies is also investigated.

Extensive simulation experiments are carried out to evaluate the performance of all delivery architectures. In particular, since the distributed locale server architecture and the distributed proxy architecture are ‘dual’ of each other, we concentrate on comparing the performance of these. From the performance evaluation, we provide recommendations on choosing suitable delivery architectures based on the server resource availability, multicast, and game’s avatar aggregation behaviors. The quantitative study in this thesis will be of benefit to future immersive voice service providers in the design of a cost effective delivery architecture for this service.

Finally, the thesis presents a resource management architecture for sharing processing resources among various real-time content creation applications including the immersive audio mixing application. Due to the inability of determining processing times for scheduling, a processing resources scheduling algorithm called Start-time Weighted Fair Queueing (SWFQ) is proposed. From analysis and simulation, it is shown that SWFQ offers good fairness and delay properties compared to current schemes. In fact, the fairness of SWFQ was comparable to Weighted Fair Queueing (WFQ) and the delay behavior is better than Start-time Fair Queueing (SFQ).

02Whole.pdf (1911 kB)

Share

COinS
 

Unless otherwise indicated, the views expressed in this thesis are those of the author and do not necessarily represent the views of the University of Wollongong.