The Observer Effect in Internet Measurement

A Case Study of Route Origin Validation

Can a network treat a well-known measurement target differently from an equivalent target that is not publicly associated with any measurement?

HotNets '26 · Mohsenpour & Pelsser · UCLouvain

•

Press “Play the story”, or use Next ▶ to advance step by step.

Speed

On the real Internet · 12,130 RIPE Atlas probes, 14 days

1,208

networks we could test reliably

drop the public target, reach the shadow targets

90.3%

their median APNIC ROV score

100%

their median RoVista ROV score

Feedback during responsible disclosure

One operator told us

They filter the public target used by Is BGP Safe Yet? on purpose: they do not wish to take part in ROV measurements that use Cloudflare-hosted prefixes.

They filtered the public target to opt out, not to improve their score.

Another network

After our disclosure, it began announcing the Is BGP Safe Yet? RPKI-invalid prefix in BGP itself, and kept doing so until the end of our observation.

Public targets can become part of operational routing configurations.

Are we measuring the Internet, or the Internet's response to being measured?

Paper Code & data