Predictive traffic management and buffer dimensioning in broadband wireless networks.

摘要

In future broadband satellite systems, onboard packet switching capabilities are expected to be implemented at the satellite. To cope with different traffic characteristics and service requirements of multimedia applications, fast packet switching with asynchronous transfer mode (ATM) has emerged as the key architecture for broadband satellite communications. Network resources such as link capacity and link buffers are dynamically assigned to cell streams from various sources by means of statistical multiplexing. Most real world traffic is bursty because most sources are transient. Predictive congestion control executed onboard the satellite is therefore needed to allow the system to operate in the optimal region of low delay and high throughput. The important features of broadband satellite systems are the limited onboard satellite buffer, the large propagation delay and low computational capabilities. We thus propose a predictive traffic management scheme that considers these limitations.; Our approach is to first propose a traffic estimator to estimate bursty traffic parameters. Based on estimates, we develop a Connection Admission Control (CAC) algorithm using the predicted cell loss ratio and saturation probability. We consider a predictive congestion control for both single class and multiclass traffic. We analyze the traffic behavior using a fluid bufferless model of a statistical multiplexing system with multiple types of traffic sources, each modeled as an “On-Off” source. We propose predictive buffer dimensioning for the onboard shared buffer of the satellite using a probabilistic burstiness curve instead of a deterministic burstiness bound. Some approximations and bounds reduce the computational complexity compared with the exact analysis of a fluid bufferless model.; For wireless local area networks (WLANs), we propose a predictive congestion control scheme which uses a Markov chain prediction method based on partial Bayesian observations of the buffer occupancy at the hub. We propose a congestion control scheme for ATM WLANs using a hub topology. The hub provides connectivity between the mobile terminals and to the ATM transport backbone. The congestion control technique is very simple to implement and incurs very low overhead on the system bandwidth (two bits per cell).