Performance of a queueing model with self-similar input traffic

Studies have shown that self-similar (fractal) processes are an accurate representation for the data traffic in today's high-speed networks. Network arrivals modeled as a Poisson process or a compound Poisson process, though presenting analytical simplicity, do not capture the scale-invariant property of data traffic. The packet inter-arrival distribution clearly differs from the exponential form being assumed and this affects buffer sizing. This paper shows through simulation studies of how the buffer overflow probability can be estimated when the arrivals are generated using a heavy-tailed distribution. A G/D/1/B queue, with a general distribution type input process, deterministic service process, single server system and buffer size B, is assumed.