Predicting User Dissatisfaction with Application Performance on Home Gateways

Home gateways connect devices in the home to the rest of the Internet. The gateway is ideally placed to adapt or control application performance to better fit the expectations of home users. However, it is not clear if we can infer user (dis)satisfaction with application performance with data available in a home gateway (i.e. mostly traffic metrics). Joumablatt et al. used machine learning methods to predict user (dis)satifaction with network application performance at end-hosts. In this paper, we study the feasibility of appying such predictors on a gateway. We show that we can measure the relevant metrics on a home gateway, we improve the proposed non-linear SVM predictor with parameter tuning, and we show that the gateway based predictor can achieve similar performance compared to the end-host predictor