Estimating Traffic Latent Due to QoS Deterioration: A Time-Series Causal Inference Approach

This paper presents a novel method for calculating the impact of Quality of Service (QoS) degradation on traffic reduction. QoS degradation often results in a decrease in Quality of Experience (QoE), prompting users to abandon applications or compelling applications to decrease content size to avoid user abandonment. In both scenarios, overall traffic is diminished. We define to the potential traffic demand before such reductions as latent traffic, which is crucial to estimate for effective capacity dimensioning and traffic engineering. Our estimation of latent traffic employs a causal inference approach, perceiving the traffic reduction due to QoS degradation as a causality from QoS degradation to traffic reduction. However, existing causal inference methods assume a one-way relationship between cause-and-effect variables. In our context, a bidirectional relationship exists, as QoS degradation is often triggered by increased traffic. Here, by assuming these two causal links operate on distinct timescales, then we can isolate them into two one-way relationships to estimate causality. Yet, the estimating of the time series is prone to both bias and variance challenges. We tackle these issues through our proposed multi-source iterative causal inference method. We validate our method's effectiveness through simulations, emphasizing the importance of temporal granularity. Moreover, our results affirm that our method can precisely gauge the latent traffic demand.