The utility model applied to layer-coded sources
Date
1998
Authors
Chen, Lei
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
In distributed multimedia systems, real-time media are streamed over a network from senders to receivers. To obtain good playback quality at the receiver end, sufficient network and system resources must be allocated to transmit and process the media streams within rigid real-time constraints. However, a major problem is: how should network bandwidth be shared among applications during real-time multimedia transport, so that contention is avoided?
This thesis presents a solution to the above problem, by combining two well established concepts: the Utility Model and layered coding and transmission. The Utility Model deals with quality of service (QoS) adaptation of multi-session multimedia systems. The QoS of individual multimedia sessions is dynamically adapted in order to achieve the system utility objective (such as maximizing server revenue), while obeying system resource constraints (total available memory, for example) and application constraints (the minimal set of resources necessary to achieve minimal acceptable QoS). With layered coding and transmission, a media source is encoded into several signal layers, each sent over a distinct network channel (such as an IP Multicast group); the receiver can pick a subset of layers to receive and decode, thus varying both the delivered media quality and the bandwidth requirement. Hence, when applied to layer-coded sources, the Utility Model should be able to decide which subset of layers to receive, based on available reception bandwidth.
To demonstrate our approach, we implemented the Utility Model and resolved issues such as QoS mapping and resource monitoring. We present a Quality Utility Extension (QUE) implementing the Utility Model in a prototype multimedia system. We describe the QUE architecture and its two activities: admission control and quality adaptation. In the QUE, QoS mappings are done by QoS Agents, per-application proxies which handle the communication with a Utility Model Engine. The QUE also deals with issues in resource monitoring and contention avoidance under the constraint of a traditional best-effort operating system. Finally, we present some implementation specifics of the QUE, along with two demo applications: VoD and QFf P.
Our experiments and results show that:
1. in a mixed (best-effort and reversion-based) system, the QUE is able to reactively adapt the QoS of applications in its domain to avoid bandwidth contention;
2. in an entirely reservation based system, the QUE respects both application and system resource constraints, while optimally adapting the QoS of applications to maximize total system utility.
The performance measurements of our implementation show that it is well-suited for real-time multimedia applications.