Open Collections

Recursive Bayesian traffic prediction for performance improvement in OBS networks

University of British Columbia

This thesis deals with traffic prediction in Optical Burst Switched (OBS) networks with self-similar traffic, i.e., traffic with long-range dependence (LRD) properties. Aggregated traffic in high speed optical networks exhibits LRD properties. OBS is a recent promising optical network technology to facilitate IP-Over-WDM (Internet Protocol Over Wavelength Division Multiplexing). To improve the quality of service (QoS) in OBS transmission networks, traffic prediction is required for dynamic resource reservation. We present and discuss a model of IP traffic based on MMPP (Markov Modulated Poisson Process), which approximates LRD traffic by mimicking the hierarchical generation of data by Internet users. The MMPP model is capable of effectively capturing the key aspects of the traffic measured on an OBS edge router, hence representing an aggregation of the traffic generated by a number of sources. The main contribution of this thesis is to derive an optimal Bayesian predictor for the burst size at the ingress router of an OBS network for MMPP approximations of LRD traffic. Bayesian prediction yields the MMSE (Minimum Mean Square Error) estimate of the burst size. As shown in a simulated OBS testbed, such Bayesian predictor can yield substantial improvement in latency reduction and service differentiation of OBS network compared to linear predictors, without substantial increasing in computational complexity.

Text

2009-12-10

For non-commercial purposes only, such as research, private study and education. Additional conditions apply, see Terms of Use https://open.library.ubc.ca/terms_of_use.

http://hdl.handle.net/2429/16331

Electrical and Computer Engineering

University of British Columbia

UBCV

DSpace