On the impact of long-range-dependent traffic in dimensioning ATM network buffer

Recent measurements of packet networks have shown that packet traffic exhibits long-range-dependent property. Studies have also discovered that this property is not adequately captured by the conventional Markov traffic models and often yields queueing behavior such as heavy-tailed or subexponential queue length distribution. These findings have led to the implication that long-range-dependent traffic has a significant impact on determining queueing performance and dimensioning buffer requirement in general. This paper examines the impact of long-range-dependent traffic on providing connectionless services over ATM networks using AAL5 protocol. As connectionless data packets enter an ATM network, they are encapsulated into AAL5 frames. Within the ATM network, AAL5 frames are segmented into cells for transmission and reassembled at intermediate connectionless servers for frame switching. A queueing model characterizing the AAL5 frame reassembly operation is developed and analyzed to investigate the impact of long-range-dependent traffic on dimensioning the frame reassembly buffer. The buffer overflow probability is analytically derived. Tight upper bound and lower bound of the AAL5 frame loss probability are also established. The analysis shows that long-range-dependent traffic and conventional Markovian traffic yield similar queueing behavior in the frame reassembly operation