Theoretical network load limit when self-similarity has no adverse effect on the network

It is well-known that the multimedia traffic is statistically self-similar, with fractal-like behavior, that is, network traffic exhibits scale-invariance at a wide range of scales. Such scale-invariance is different from traditional models of network traffic. It is also well-known that self-similarity degrades the network performance by requiring large queueing buffers, causing delay and packet dropping problems, and that the traditional queueing theory is inadequate to predict network performance. The conventional wisdom is that the higher the load on the network, the higher the degree of self-similarity. In this paper we first show a method of how to approximate a queueing buffer size with a self-similar input process, then we derive a theoretical network load limit when self-similarity has no adverse effect on the network. This load limit is found to be about the same load value published in literature, on average delay, with actual Ethernet traffic traces compared against curves obtained from the traditional queueing network analyser