Measurement-based traffic management for QoS guarantee in multi-service networks.

摘要

This dissertation addresses some fundamental issues of measurement-based traffic management in multi-service networks. The major concerns include proper traffic measurement time interval and online QoS monitoring, the predictive network control performance, and the scalable scheme for weighted fairness guarantee in a DiffServ Assured Forwarding (AF) environment. The purpose is to provide the strict or elastic Qualify of Service (QoS) over multiple time scales at a multi-service ATM switch or a next-generation Internet router.; Given the practically ambiguous or insufficient network traffic knowledge, network operator relies on the on-line or off-line measurement of traffic statistics and resource utilization, in addition to a reliable traffic model, for robust control actions. Naturally the key points are how to accurately measure the real-time traffic bandwidth demand and how to guarantee the QoS over both short- and long-term periods. To answer these questions, we proposed a Virtual Queue (VQ) scheme to adapt the critical measurement interval T_c to real-time traffic dynamics and online measured QoS. The T_c derives an accurate load estimate at a single finite-buffer network node under explicit cell/delay loss constraints.; Assume that our measurement scheme has captured the accurate traffic load over the concerned time scales. However, it only reflects the past. We may set up certain traffic model to predict the traffic future for rate-based network controls. So what is the remotest future an online predictor can forecast confidently? This question motivated the predictability analysis of network traffic based on the idealized a priori traffic knowledge: stationarity, ergodicity and precise models. Our study reveals the typical MPI, i.e., the maximum prediction interval, the predictive control efficiency, and their tradeoff.; Due to the actual demand from the current industry, network researchers are working on the traffic management schemes for the next generation data network offering a variety of QoS guarantee. The merged DiffServ and MPLS technology in the Internet is a promising future, where heterogeneous flows ask for scalable weighted fairness from their AF services as a minimum QoS requirement. To meet this demand, we proposed a network-service model—SCALE-WFS. The model integrates the labeling concept from MPLS/ATM and the aggregated service concept from DiffServ with reliable traffic measurement. Performance evaluation shows its advantages over the popular schemes such as the Random Early Discarding (RED) and the core-stateless fair queueing (CSFQ). Again the traffic measurement plays a crucial role in the highly unpredictable and dynamic environment.