Summary

  • In 2005, Gaurab Raj Upadhaya and Vijay K. Adhikari tested a simple claim: if a Tier-1 network never buys transit, no ordinary AS path should need more than two Tier-1 candidates.
  • Their 2006 work with Bill Woodcock found paths with three or more names from the test list, plus recurring oddities. They asked operators to explain the paths; the observations alone did not disclose private contracts.

The troublesome route had room for too many kings. In the APNIC 22 slides, a path towards 65.215.36.0/24 placed several supposed top-tier networks around Cybersites. The team's proposition allowed at most two transit-free networks in one normal AS path, meeting at a peering center. Three names from the candidate list meant the model, the configuration or the boast needed another look. A marketing title had reached the part of the Internet where routers keep receipts.

Upadhaya did not begin with a universal registry of prestige. His 2005 APNIC 20 presentation, with Vijay K. Adhikari, used Route-Views archives and a Tokyo transit-router view. They selected frequent transit ASNs in each view, intersected the lists and tested the resulting candidates against observed paths. The slides say plainly that there is no single global routing table. They also warn that content providers were underrepresented. The test could locate awkward paths; it could not certify every commercial relationship on Earth.

The earlier exercise had already made geography a nuisance for the label. Removing AS4637, Reach, from the candidate list cut the global exceptions, while the Japan view alone did not produce the same exceptions. The authors treated that as evidence that a network might sit at the top of a regional graph without occupying the same place everywhere. Their slide called “Tier 1” a socioeconomic concept. That is an unusually honest caption for a routing diagram: the market sells one badge; the network gives you several vantage points.

By APNIC 22 in September 2006, the team had widened the observation. Upadhaya, Woodcock and Adhikari showed paths with three or more candidate Tier-1 ASNs and sorted other oddities into inconsistent origin ASNs, non-contiguous repeats, private-ASN leaks and unallocated-ASN leaks. They also drew “X” relationships, where two ASNs appeared to announce through each other. One apparent puzzle involved AS9498 and AS9730, which the slides identify as parts of the same company. A strange path can expose a real leak, a corporate structure, a measurement blind spot or a bad assumption. The diagram does not choose for you.

The meeting transcript catches the work at its most useful moment. Upadhaya described Route-Views, PCH collectors, a Japanese view and feeds from Europe; he was still looking for a South American feed. Philip Smith offered a possible explanation for a Tiscali case. Upadhaya pushed back on the export logic and asked Asia-Pacific providers for input. The presentation's proposed email alerts and web portal were plans, not a delivered product. The evidence had become sharp enough to ask an operator a specific question, and still incomplete enough to need the operator's answer.

That is the wager behind this old study. A carrier can buy a rank in a reader's mind by saying “Tier 1.” Upadhaya's team made the rank carry a testable claim about transit and path shape. When the path breaks the claim, the next move is neither applause nor accusation. It is to identify the prefix, collector, time, AS sequence and the network that can explain it. Prestige is cheap until somebody opens the routing table.