Part of Advances in Neural Information Processing Systems 9 (NIPS 1996)
Timothy Brown
This paper presents a method that decides which combinations of traffic can be accepted on a packet data link, so that quality of service (QoS) constraints can be met. The method uses samples of QoS results at dif(cid:173) ferent load conditions to build a neural network decision function. Pre(cid:173) vious similar approaches to the problem have a significant bias. This bias is likely to occur in any real system and results in accepting loads that miss QoS targets by orders of magnitude. Preprocessing the data to either remove the bias or provide a confidence level, the method was applied to sources based on difficult-to-analyze ethernet data traces. With this data, the method produces an accurate access control function that dramatically outperforms analytic alternatives. Interestingly, the results depend on throwing away more than 99% of the data.