Heuristic rule based neuro-fuzzy approach for adaptive buffer management for Internet-based computing

The problem of buffer management in an Internet computing environment is concerned with effective determination of an appropriate buffer size so that retransmissions in message passing can be reduced or even avoided. Such retransmissions can cause significant time delay in the message being delivered because of buffer overflow at the reception side. In our previous work, two buffer management approaches were proposed, namely, the P+D and the P+I+D schemes. These two methods are only based on partial information on the system and the rule form lacks adaptive power. This paper aims at developing a more powerful model for adaptive buffer management. Based on the idea of heuristic knowledge, we propose a novel fuzzy model with a modified consequent part of the Takagi-Sugeno type fuzzy inference approach. The use of the prior knowledge In the connectionist fuzzy system improves the reliability of buffer size prediction, and enhances the capability of domain knowledge representation and comprehension. A successive online learning algorithm is outlined using convergence algorithm and gradient descent technique. Simulations of the proposed model confirm that neuro-fuzzy approach is indeed a better adaptive buffer management solution for preventing message loss due to overflow.