The Markov-modulated Bernoulli process (MMBP) is widely used in modell
ing traffic sources. The authors propose a so-called high-low (H/L) mo
del, a two-state Markov chain, to characterise multimedia traffic on A
TM-based broadband networks. The H/L model has two nonzero-valued stat
es corresponding to the high-activity and the low-activity level of mu
ltimedia traffic. The queue-length distribution of an ATM switch fed w
ith N identical MMBP sources was derived using a probability generatin
g function coupled with a matrix geometric approach. By decomposing ma
trices into their spectral representation forms the queue length can b
e evaluated by considering the roots of its characteristic function. N
umerical results for the case of two MMBPs are also presented.