Markovian and self-similar data traffic models are currently being proposed as accurate characterizations of the way data traffic behaves on an asynchronous transfer mode (ATM) network. Because of the dearth of traffic data from real ATM networks, researchers in this area have relied on simulated data and data from Ethernet networks. Ethernet and ATM are both packet technologies and share some similarities. There are, however, important differences between the two technologies which largely preclude the use of Ethernet traffic data in studies of ATM network traffic behavior. We propose a model for ATM data traffic based on analysis of traffic traces from two real ATM networks. This analysis reveals that the interarrival time distribution for data traffic is bimodal, in a similar fashion to Ethernet, but is a mostly lognormal mixture distribution with a very small self-similar or pseudo self-similar portion in the right tail
