Summary

  • RIPE Labs' original investigation found old equipment treating 128.0.0.0/16 as special address space. RIPE Atlas measurements showed the data-plane consequence, and Kühne's 2012 account records outreach to 121 operators and a rise in observed probe reachability from roughly 70% to 85%.
  • In the 2013 follow-up, more than 95% of probes in a bounded two-network test reached 128/16 destinations. Reachable control targets in the same networks helped distinguish prefix filtering from an ordinary failed ping.

The address had paperwork. The routers had old memories. Some Juniper software still blocked 128.0.0.0/16 as a “martian” prefix, the RIPE Labs investigation reported, even after the old special-purpose classification ceased to apply. A network receiving an allocation from the block could possess perfectly ordinary addresses and still lose traffic at somebody else's filter.

The RIPE NCC made the problem measurable. RIPE Atlas probes tested whether packets could reach a target in the prefix and come back. That data-plane view supplied names of networks where the journey broke. It also supplied a question more useful than a general plea to update routers: which operator owns the filter in this path?

Kühne's June 2012 report describes a team effort with contributor Kjell Leknes. They contacted 121 operators covering the observed filtering cases. Thirty-seven already knew about the problem, fifteen said they did not, and many offered no reply. Thirty-four told the team they would change filters within a month. Roughly two months into the effort, 47 operators had stopped filtering the prefix. In the test population, the share of probes reaching the target moved from about 70% to about 85%.

The method changed during the campaign. The team tried calls and emails, then learned that operators with clear public NOC addresses generally preferred email. Smaller operators without a clear contact got a call. It is a wonderfully unglamorous lesson in Internet repair: a route can be visible in a chart, but someone still has to find the right inbox and explain why the rule is old.

The follow-up came with better controls. In February 2013, Kühne reported two sets of about a thousand Atlas probes testing addresses in two different recipient networks. Each network offered an established reachable target as a comparison. A probe that could reach its control but failed on 128/16 supplied a sharper signal than a lone failed ping. More than 95% of probes reached the 128/16 targets in that test; 64 failed to get a reply from a documented target. The remaining failures still mattered to the address holders trying to run services.

Kühne credited Kjell Leknes and Rene Wilhelm in that later work. The operators who changed their filters supplied the real network effect. The story's punchline belongs to no magic command: an address moved from “martian” to usable through a sequence of observation, contact, local decisions and retesting. The old rule had survived in equipment longer than it survived on paper.