Smart Experts for Network State Estimation

Several network protocols, services, and applications adjust their operation dynamically based on current network conditions. Consequently, keeping accurate estimates of the network and its performance as it fluctuates over time is critical. For example, both TCP and IEEE 802.11 periodically adapt some of their key operating parameters, namely, the retransmission timeout and the contention window size based on the average round trip time and the number of collisions, respectively. In this paper, we present a novel mechanism to estimate "nearfuture" network performance based on past network conditions. We call our approach to network performance estimation as smart experts for network state estimation (SENSE). SENSE uses a simple, yet effective, algorithm combining a machine learning method known as fixed-share with exponentially weighted moving average (EWMA). SENSE also introduces novel techniques that improve the predictability of the fixed-share framework without increasing computational complexity. SENSE is thus able to respond to network dynamics at different time scales, i. e., long-and medium-term fluctuations as well as short-lived variations. We evaluate SENSE's performance using synthetic and real datasets. Our experimental results show that, when compared to fixed-share and EWMA, SENSE yields higher estimation accuracy for all datasets due to its ability to more closely track data fluctuations.