Queue response to input correlation functions: discrete spectral analysis

A new concept of spectral characterization of the wideband input process in high-speed networks is examined. The eigenstructure technique, through the modeling of input Markov chains, helps localize wideband sources in a subspace, especially in a low frequency band. Simple periodic chains are used for the construction of the input rate process. The input power spectral distribution is defined in a discrete frequency domain. Each input traffic stream is characterized by an independent Markov modulated Poisson process (MMPP). The underlying Markov chain is used to reflect the time autocorrelation properties of the input process at a macro level. Expressions are derived for correlation and power spectral functions of the MMPP input. A queuing analysis with multiple periodic input functions is described. The queue response to input correlation functions is examined. The advantages of using the spectral domain to analyze and design network control and resource management are discussed