Article ID | Journal | Published Year | Pages | File Type |
---|---|---|---|---|
475385 | Computers & Operations Research | 2008 | 14 Pages |
Computing or estimating link blocking probabilities is a fundamental ingredient in network design and engineering. While in traditional telephone networks this was easily done by Erlang's formula, it became much harder in today's complex networks that carry very heterogeneous traffic. We present a simple, efficient method to estimate the blocking probability and link utilization for general multirate, heterogeneous traffic, where the individual bandwidth demands may aggregate in complex, nonlinear ways. The estimation is derived without adopting conventional performance modeling assumptions, such as Poisson arrivals or exponential holding times, thus allowing non-standard behaviour patterns, including special phenomena of IP networks, such as self-similarity. Despite its generality, our estimation maintains an optimally tight exponent and is capable of handling apparently very different cases in a unified way. The approach also makes it possible to make estimations under incomplete information. We show that the results are easily applicable for fast, robust link dimensioning, especially in case of complex traffic patterns, under partial information. Moreover, it is very well fitted for embedding into network level optimization tasks, due in part to simplicity and in part to convexity properties.