Endpoint support
We used Scamper to measure path behavior between our vantage points and all endpoint datasets, with the goal of identifying ECN bleaching along network paths.
Results on network support for ECN
We characterized infrastructural support for ECN by inspecting whether routers along the network path between clients and servers or other Internet endpoints correctly preserve ECN-related ToS bits. We ran Scamper’s Paris traceroute from all vantage points to all targets in our datasets, to detect potential bit mangling of ECN-related ToS bits—ECT(0), ECT(1), and CE. We also ran traceroutes without any ECN flags for comparison. From each vantage point, we conducted four traceroutes per target. We additionally traced to one random address per prefix in the BGP prefix and registry datasets
From Vantage Points to All Endpoints
-
To assess ECN bleaching, we analyzed more than 52 million successful IPv4 and IPv6 path traces collected with Scamper across the Tranco, university, and IPv6 Hitlist datasets. As shown in Fig. 1, ECN markings were preserved along the vast majority of paths, with bleaching observed in only 4.3% of traces overall. Bleaching was more common in IPv4 than IPv6, indicating that ECN-related IP header fields are generally preserved by modern network infrastructure. Further analysis revealed that a small number of networks accounted for a disproportionate share of bleaching events, with resetting ECN-related bits to zero being the most common behavior.
a. IPv4.
b. IPv6.
Figure 1. Routers or middleboxes bleach ECN-related ToS bits in only about 10% of paths to targets in the Tranco, university, and IPv6 Hitlist datasets.
From Vantage Points to Arbitrary Prefixes
-
We extended the bleaching analysis beyond web servers by probing random addresses drawn from routable IPv4 and IPv6 prefixes. Consistent with the results for server infrastructure, ECN markings were preserved on more than 90% of successful paths, indicating that ECN bleaching remains relatively uncommon in the broader Internet. When bleaching occurred, it was typically observed closer to the destination network, and resetting ECN-related bits to zero was the dominant behavior. Unlike the server datasets, no single AS or network emerged as a major contributor to bleaching events. Additional details are provided in the paper.
a. IPv4.
b. IPv6.
Figure 1. Routers or middleboxes bleach ECN-related ToS bits in only about 10% of paths to targets in random addresses drawn from routable IPv4 and IPv6 prefixes.