Accuracy Improvement for Spatial Composition-Based End-to-End Network Measurement

Spatial composition method for end-to-end network measurement avoids lengthy measurements of a path and the path is divided into some sub-paths when multiple measured paths share the underlay network route. The performance of the overall path is estimated by spatially composing measurement results of the sub-paths. This method is quite effective for measurement in overlay networks where many paths between overlay nodes share the underlay network. However, such estimation methods may include the additional errors caused by the measurement inaccuracy of sub-paths. Therefore, we need to assess the estimation accuracy of the spatial composition-based method and introduce statistical processing to suppress the estimation errors. In this paper, we propose statistical processing methods of measurement results to improve estimation accuracy of spatial composition-based measurement method for packet loss ratio. We introduce a statistical test for measurement results to exclude outliers from spatial composition. We also propose some statistical indexes for determining whether we should discard the measurement results and reconduct the measurement. We evaluate the performance of the proposed method by using measurement results obtained on Planet Lab environment. From the evaluation results we find that proposed two methods can decrease the estimation error of the spatial composition of packet loss ratio by 36% and 23%, respectively.