| Dynamic Bandwidth Allocation |
|
|
While a significant portion of Internet traffic is still best-effort, emerging applications with different quality of service(QoS) requirements (e.g., multimedia) are evolving in a faster pace and require more than best-effort services. The Integrated Services (IntServ) model attempts to achieve an end-to-end per flow QoS through RSVP, and it lacks scalability. The differentiated services (DiffServ) model, on the other hand, attempts to classify packets based on their QoS requirements to receive different services, and it is the focus of recent IETF activity. To ensure QoS, the admission control policy and the packet scheduler should work in tandem to i) control the misbehaved traffic through policing, shaping, metering, pricing techniques, and ii) contain the delay, delay variation, packet loss rate, and bandwidth loss through robust packet scheduling policies. In this paper, we propose an adaptive admission control that reduces bandwidth loss when the traffic is light and enforces a stringent traffic regulation when the traffic load is high. A similar mechanism is used for the packet scheduler that maintains the QoS metrics adaptively within their limited boundaries. Internet2 traffic traces and Opnet synthetic traffic have be used to evaluate the performance of the algorithms. The exponential growth of the Internet has put a two-fold demand on switches and IP routers; higher bandwidth and lower latency. In addition, new services that support multicast and multimedia applications require sustained bandwidth and contained delay and delay variation (jitter) to support their quality of service (QoS) requirements. While for some applications a timely packet delivery is of the utmost importance (e.g., real-time video), for some other applications a reliability of a packet delivery is far more important (e.g., e-commerce). Current IP networks provide best-effort services with minimal QoS provisioning. |
|
| Last Updated ( Sunday, 17 October 2004 ) |