Summary

  • William B. Norton’s early peering work treats traffic analysis as a way to find possible counterparts, not as a universal rule for deciding what an interconnection is worth.
  • His 2005 discussion of peering ratios records arguments on both sides and remains a draft he asked colleagues to review. The distinction still matters: a measured ratio is evidence; a threshold is a policy choice.

The number is not the bargain

Imagine two networks exchanging traffic. One says most of the bytes arrive from its counterpart; the other says that the same relationship saves transit fees, shortens paths to valued routes or reaches customers it could not efficiently serve another way. Both can report traffic accurately and still disagree about what the exchange is worth.

A ratio is tidy because it reduces a complicated relationship to a comparison. But the number depends on what was counted, where the boundary was drawn and when the measurement was taken. Even a correctly measured imbalance does not settle who carries more cost, which routes are difficult to replace, how many users benefit, or whether a connection is cheaper than the alternatives. It can also become an eligibility test: a network that misses a stated ratio may not get to negotiate the very relationship whose value is in dispute.

That distinction runs through William B. Norton’s work on Internet peering. The public record does not make him the inventor of peering or of ratio policies. It does show him returning to a practical question: which evidence helps two networks decide whether to connect, and which evidence is being asked to do more than it can?

A decision process, not a magic threshold

An early Norton paper, preserved as a draft and later catalogued by CAIDA as an Equinix white paper from 2001, divides peering into three phases. First, traffic-engineering information helps identify likely counterparties. Next come contact, qualification and negotiation. Only after the parties decide that a relationship is worthwhile do they work through implementation.

The order matters. Traffic can help an operator find where a direct path might reduce purchased transit or improve performance. That makes traffic volume useful in the first phase. It does not make a single inbound-to-outbound ratio a neutral verdict in the second. Qualification involves each network’s own costs, alternatives, policy and expected benefit. Implementation then raises separate questions about locations and technical arrangements.

The paper’s historical calculations are not current prices, and its process is not proof that every ISP follows the same sequence. Its lasting analytical value is narrower: measurement can point toward a conversation without deciding what the counterpart must accept.

Norton’s public roles placed him near both the operational and commercial sides of this process. He chaired NANOG in the 1990s and later worked as Equinix’s co-founder and chief technical liaison, according to his own biography and contemporaneous NANOG records. In a 2004 Peering BOF at NANOG 30, he moderated a debate over restrictive peering policies. The meeting notes preserve competing cases: one side stressed cost, scale and revenue; the other warned that restrictive rules could deter useful relationships. The record shows debate, not a vote conferring authority on the room.

What the 2005 ratio debate could—and could not—prove

The following year, after a debate at Peering BOF X during NANOG 35, Norton posted a draft called “The Folly of Peering Ratios” to the mailing list. In that message he said that some of the strongest points had emerged in the Q&A and informal conversations that followed the formal debate. He grouped roughly six arguments and counterarguments in a version 0.5 paper, then asked peers to review whether he had represented the issue accurately before circulating it more broadly.

The arguments were not all the same claim. Some operators objected to carrying large volumes of another network’s content. Others focused on the distance or distribution of traffic, the effect of a new peer on ratios with existing peers, revenue, capacity constraints, or a desire to limit additional connections. Norton’s draft argued that a ratio was a weak proxy for the value received and that a network-load problem could have other remedies. Those are arguments he advanced in a draft—not a measured finding that every ratio is misleading or every restrictive policy irrational.

His 2014 Internet Peering Playbook makes the limit more explicit. In its chapter on Internet peering, Norton says there is no standard way to calculate the absolute value of a peering relationship. Possible measures might include traffic volume, the desirability or uniqueness of routes, or the number of people reached. Those dimensions can diverge. A smaller flow may reach an otherwise unavailable destination; a larger flow may be inexpensive to carry; a route may have little value if it duplicates alternatives.

The point is not that traffic should be ignored. It is that a byte count and a price are different objects. A ratio might be a useful screen under a particular policy. It does not become an objective definition of fairness merely because it is numerical.

A public argument is not a mandate

Norton’s contribution is best understood as documentation and translation: he turned operator conversations into a staged decision model, public papers and teaching exercises. The Internet Society later described his peering simulation game as a way to help participants understand peering economics and end-user benefits. That reach matters, but it does not establish that his writing speaks for every operator, or that any one forum can settle whose costs count.

The distinction is also a question of authority. Traffic records describe activity at an observed boundary. Operators decide which costs, routes and customers enter their business case. Each network sets its own peering rules and can change them. People who speak in a meeting contribute evidence and expertise; attendance alone does not delegate the power to represent every affected network. Nor does an aggregate ratio tell us which party had the power to define the denominator.

A defensible policy would therefore say more than “the ratio is out of range.” It would state the measurement window and traffic scope, explain what decision the ratio is meant to inform, distinguish observed load from economic value, identify which alternatives were considered, and disclose who can grant exceptions or revise the rule. That record would not force networks to peer. It would make clear whether the number is being used to locate a candidate, estimate a cost, or exclude a counterparty.

Norton’s papers do not provide a universal replacement formula, and the archive does not show that his draft ended the argument. Their useful boundary is more durable: traffic evidence can begin a negotiation, but it cannot silently stand in for the bargain.

Sources