An investigation of measurement-based load and QoS management in a distributed audio environment.

Abstract

Recently, Massively Multiplayer Online Games (MMOGs) have received significant popularity. Made up of populations of physically distributed players, MMOGs provide user immersion into the playing environment through realistic graphics and sound. However, many lack real-time communication between users, relying upon text-based and simple VoIP systems. The Immersive Audio Communications for Massively Multiplayer Games (IACMMG) and Dense Interactive Communications Environment (DICE) projects, currently being developed by the Smart Internet Technologies CRC, aim to change this by providing realistic two dimensional voice communications. As MMOGs cater for a global audience, the projects are designed to utilise a network of audio processing servers to which users are connected for the purposes of communication. These projects face a number of challenges, two of which we address in this thesis. The first is concerned with the finite resources of each server and is thus a question of how to effectively balance user load over the available audio servers. The second problem is concerned with ensuring that users receive acceptable audio Quality of Service (QoS). In order to address the first problem, a load balancing architecture is proposed. Four load balancing algorithms are designed and implemented and their respective trengths and weaknesses explored. To address the second problem, we explore how QoS should be defined in an MMOG context. The IACMMG and DICE projects are VoIP based, thus QoS is primarily dependent upon packet delay and loss. We explore techniques for information gathering from packet traffic and use it for equating perceived QoS. This leads to the development of a QoS algorithm designed to measure the QoS of conversations within the virtual environment of the MMOG, reacting to the perceived level of QoS, and moving users to servers capable of providing higher levels of QoS should this be necessary. The load balancing and QoS aspects of the IACMMG and DICE projects are dependent upon one another, and thus we are left with a problem of two potentially conflicting objectives. To address this we integrate the two architectures, modifying each to make it compatible with the other. The culmination is a complete architecture capable of both load and QoS management.