Measurement-based traffic characteristic estimation for QoS-oriented IP networks

In order to properly exploit future business opportunities by offering IP networks to users, network operators and service providers, the performance of the WANs needs to be of high quality. In any broadband WAN quality of service (QoS) is only guaranteed by ensuring that the underlying network performance, in terms of information loss and delay, is meeting the required levels. This means that to provide services for heterogeneous applications with guaranteed QoS, sufficient bandwidth and buffer space must be allocated in the network. It also means that different classes of application (delay sensitive, loss sensitive etc.) must be partitioned sensibly. To ensure correct partitioning, the IETF proposed DiffServ to overcome the scalability problem that existed in RSVP. In this paper, we consider the scenario of voice traffic being carried in a DiffServ domain with Class-Based Weighted Fair Queuing. We have proposed a novel bandwidth estimation mechanism oriented to ensuring that the VoIP traffic meets its QoS requirements. This estimation method is based on traffic measurement and queue length observation, and we illustrate how to carry out measurements in order to capture the traffic characteristics. Our methodology consists of aggregate traffic measurements, utilization and queue length observations, and based on this data, we can deduce the traffic characteristics. Once the traffic characteristics are captured, we can then estimate the bandwidth necessary to meet the QoS requirements, and check to see if the existing bandwidth allocated is adequate. We further present simulation results which demonstrate that this scheme is very promising.