Summary

  • Susan Estrada's CERFnet paired early commercial reach with customer accounting reports, 24-hour monitoring and support practices that made a regional network more legible as a service.
  • The durable lesson is not that a packet count equals a bill. Reachability, observation, accountability, charging and usefulness are separate surfaces, and a credible provider has to say where each begins and ends.

Imagine the artifact on a customer's desk. It is not a map of the Internet and not a photograph of a router. It is a report covering a period of service: traffic observed, interruptions classified, perhaps a note about what the provider could and could not see. The pages do not create connectivity. They make a claim about it inspectable.

That distinction helps recover Susan Estrada's place in the commercial Internet without turning the story into a parade of corporate names. The Internet Hall of Fame says she founded CERFnet in 1988, taking an initial $2.8 million National Science Foundation award and growing a California regional IP network from 25 sites to hundreds. It served academic and commercial communities. The same profile highlights dial-up IP, accounting reports for customers and 24/7 monitoring among CERFnet's notable early practices.

Those three items belong together. Dial-up IP changed the form in which a customer could reach the network. Continuous monitoring changed the provider's ability to notice and investigate service state. Accounting reports changed what could be discussed across the provider-customer boundary. None was sufficient alone. Together they pushed a research-network attachment toward something an institution could buy, operate and contest.

“Accounting” needs care. In ordinary speech it suggests an invoice. In the Internet engineering record it is broader. RFC 1272, published in 1991, describes usage reporting for service providers and consumers, discusses meters at administrative boundaries and says observations kept by different parties should be reconcilable. The functions may inform policy, cost allocation, quotas or charging. A measured packet is not automatically a priced packet, and a usage report is not necessarily a bill.

This matters because charging can create false precision. A counter can say that bytes crossed a boundary. It cannot say that an application was useful, that a delay occurred inside the provider's responsibility, or that the institution's staff received effective help. Even the placement of the meter expresses a governance choice: which administrative boundary matters, whose traffic becomes visible and which failure remains outside the frame.

The surviving CERFnet presentation in the proceedings of the 21st IETF meeting offers a telling piece of evidence. Presented by Estrada, it includes charts that classify outages by type and duration across two periods. The scan is too degraded to support confident reconstruction of every number. Its value is institutional: the network brought an operating record into a public technical forum. It showed that service history, not only topology, belonged in the conversation.

An outage chart can still mislead. Short incidents may be severe; a long event may lie beyond the provider's control. Scheduled maintenance, a local circuit, a router error and a backbone failure are not interchangeable. A customer needs the category definitions, observation point, time window, exclusions and restoration state. Without those fields, a chart is visibility without accountability.

Estrada also entered the Internet Engineering Steering Group from the perspective of a mid-level operator. The same IETF proceedings identify her as a co-director of the Operational Requirements Area, whose work included user connectivity, operational statistics, joint management and benchmarking. Its report warned that new providers were entering the market and that operations staff had to bring real requirements back to protocol developers.

That movement—from production into standards work—is central to the story. Protocols make interconnection possible, but operators discover whether procedures are affordable, whether a transition gives adequate notice, and whether the evidence produced during a fault can support coordination. Estrada's 1991 FARNET presentation put the point plainly: member networks were production environments with economic and operational constraints, and new standards had to be workable as well as technically coherent.

FARNET itself represented another form of legibility. The presentation describes a federation spanning national, regional, state, provincial and local providers and serving more than 1,000 organisations. Its stated work included improving customer and network information services, advocacy, programme activity and group discounts. A production Internet was not merely a mesh of autonomous routers. It was also a set of organisations pooling information, negotiating costs and carrying the consequences of change.

The policy environment was shifting at the same time. The Internet Society's historical account says NSF encouraged regional networks to seek commercial customers and use expanded facilities to reduce subscription costs, while the NSFNET backbone's acceptable-use policy constrained commercial traffic on national transport. Private long-haul networks grew around that boundary. Estrada's work with PSINet and UUnet on the Commercial Internet Exchange belongs in this wider transition, but the exchange did not by itself create a dependable retail service.

Interconnection answers one question: can traffic pass between administrative domains under the agreed arrangement? A customer report answers another: what happened to the service attributed to this relationship during this period? Support answers a third: who owns the next action when the customer cannot use what the topology says is available? Commercial acceptance requires all three questions to have credible answers.

RFC 1167, a 1990 discussion of a national research and education network, anticipated the coordination burden. It called for agreements on monitoring and management, internetwork accounting and exchange payments, and problem identification, tracking, escalation and resolution. It did not prescribe CERFnet's system. It captured the environment in which multiple providers would have to connect technical evidence to organisational responsibility.

The customer's own institution carried duties too. RFC 1359 told prospective connecting organisations to budget beyond the physical line: campus networking, policy, training, documentation, security, promotion, support and continuing operations. It described monitoring usage patterns and known problems, while also warning that clean statistics could not overrule unhappy users. A connection became a service only through work on both sides of the demarcation.

This creates an asymmetry that persists today. The provider sees backbone state that the customer cannot. The customer sees local failures, user confusion and unmet needs that the provider's monitor cannot. Each party can produce a technically accurate account that omits the other's experience. A useful service record therefore needs reconciliable observations and an explicit dispute route, not a single omniscient dashboard.

Estrada's 1993 buyer's guide fits this operating view. The Hall of Fame profile describes it as practical advice for obtaining Internet service. A buyer's guide assumes comparison: service boundaries, responsibilities, costs and support must be expressible well enough for an institution to choose. The commercial Internet became plausible not only when sellers appeared, but when a buyer could ask informed questions of them.

The appropriate historical claim is modest. CERFnet did not invent every component of network operations or accounting. The available sources do not establish its prices, remedies, profitability or customer satisfaction. They do show Estrada treating measurement, information services and operational coordination as part of the product. That is more consequential than a slogan about privatisation because it identifies the machinery through which a market becomes governable.

A contemporary service receipt can extend that machinery. It can state the customer and service boundary, reporting interval, observations, exclusions, incidents, restoration evidence, responsible party and escalation path. It should separate usage from price and provider measurement from user outcome. The goal is not maximal telemetry. It is enough shared evidence to decide what happened and who must act next.

The network on Estrada's desk was becoming commercial at the moment it was learning to explain itself. Its packets crossed new boundaries, but its reports created a different connection: one between technical state and institutional responsibility. A market can sell reachability without that connection. It cannot sustain trust for long.

Sources