A Light-Weight Statistical Latency Measurement Platform at Scale

The statistical values of the latencies between two sets of hosts over a given period, which is referred as to statistical latency, can benefit many applications in the next-generation networks, such as Network in a Box (NIB) based resource provisioning. However, existing methods can hardly achieve low measurement cost and high prediction accuracy simultaneously in large-scale scenarios. In this paper, we design a light-weight statistical latency prediction platform named DMS, which leverages the existing open recursive domain name servers in the network to reduce the measurement cost and takes advantage of a metric space to guarantee the prediction accuracy. Theoretical analysis shows that the predicted latency is statistically approximate to the actual value. To evaluate the performance of DMS, we implemented a prototype system in Mainland China. Compared to existing methods, DMS can reduce the relative error by 18.5% for realtime end-to-end latency prediction and 33% for statistical latency prediction.