Robust ABR control for uncertainties in long-range dependent traffic

Network performance, for the engineer designing traffic management protocols, is the key issue as it ensures that the network is functioning within its safe operating limits and that users of various applications are satisfied with the service received. Thus, in the face of uncertainty regarding the true nature of the system and traffic traversing the system, robustness entails maximising the performance of the network across the entire range of possibilities. We investigate standard robust prediction techniques for the design of ABR rate control mechanisms, but, while these optimize for the case with the greatest uncertainty, the system performance can suffer significantly for other cases. By introducing a spectral density comparison term to the entropy, we arrive at a performance measure which achieves robustness in the sense of maintaining high performance under uncertainty. Performance also includes computational efficiency, and various long-range dependent models are compared with regards to the robustness of the system, allowing the model with the least computational expense to be selected.