Causal role of ECMP routing in prior censorship measurement variance

Ascertain whether Equal-cost Multi-path (ECMP) routing is the sole cause or a contributing cause of the country- and ISP-level inconsistencies reported in prior remote (outside-in) Internet censorship measurement studies, and quantify the extent of ECMP routing’s causal influence across DNS, HTTP, and HTTPS protocols and diverse national network environments.

Background

The paper documents pervasive variance in censorship measurements observed in prior studies across DNS, HTTP, and HTTPS, including non-uniform behavior at country and ISP levels. It then demonstrates that ECMP routing can significantly influence observed censorship outcomes and explains several mechanisms by which such differences arise (e.g., routing through failed/misconfigured censoring nodes, geographic path diversity, and differing AS paths).

Despite these findings, the authors explicitly note they cannot conclude that ECMP routing is the sole or a contributing cause of all previously reported measurement inconsistencies. Establishing the precise causal contribution of ECMP routing to past observed variations remains unresolved, motivating targeted causal analyses that differentiate ECMP-induced effects from other factors such as geography, ISP policies, and censor load.