Summary

  • John S. Quarterman helped turn a collection of unlike networks into something people could compare by naming the population first and mapping it second.
  • His maps and measurements were useful precisely because they were partial: service definitions, probes, sampling windows and geographic assumptions set the limit of every result.

The most persuasive part of a network map is also its most dangerous feature. It converts an invisible, changing system into a stable surface. A line appears to join two places. A bar appears to measure a country. A rising curve appears to settle how large the Internet is. The eye receives an answer before it asks what was observed.

Quarterman spent much of the Internet’s early public expansion working on that prior question. His 1990 book The Matrix: Computer Networks and Conferencing Systems Worldwide did not describe one uniform infrastructure. RFC 1175’s contemporary bibliography records a 746-page survey organized geographically, with networks, gateways, address syntax, protocols and practical access information. The book’s object was a world in which systems could communicate through different technical and administrative arrangements. The map had to preserve those differences before it could claim a whole.

That is why “the Matrix” mattered as more than a dramatic name. Quarterman used it for the wider set of computer systems able to exchange electronic mail. The Internet was an important part of that set, but mail reachability did not necessarily provide interactive IP services. BITNET, UUCP, FidoNet, commercial mail systems and the TCP/IP Internet could share messages without becoming identical networks. A boundary based on email produced one population; a boundary based on interactive services produced another.

The distinction became an explicit measurement doctrine in RFC 1935, written by Quarterman and Smoot Carl-Mitchell and published as an Informational memo in 1996. The document starts from disagreement over what belongs inside the Internet. It then proposes nested categories: a supplier-capable core able to offer interactive services, a consumer-capable Internet that may sit behind firewalls, and the larger mail-reachable Matrix. These were chosen analytical categories, not natural borders waiting to be discovered.

One example in the RFC makes the politics of observation visible. Russia could be reached from the Commercial Internet Exchange and other networks for a period in 1992, but not from NSFNET. An observer who treated NSFNET reachability as the test could describe Russia as outside. Another observer, standing elsewhere, could see a connection. The packets did not change when the definition changed; the reported Internet did.

Quarterman and Carl-Mitchell therefore resisted using “connected” or “reachable” without a service and a context. A machine may accept mail but not an interactive session. A consumer behind a firewall may retrieve information but not distribute it. A named host may not correspond neatly to one physical computer or one active user. The vocabulary was not pedantry. It was the measurement contract that made later counts comparable.

MIDS—Matrix Information and Directory Services, the organisation Quarterman founded—carried that contract into visual work. Its charts and Matrix Maps tried to show growth, geography and network performance at a scale that prose could not hold in working memory. The achievement was not simply the drawing. It was the decision to keep several measures separate instead of compressing them into one impressive number.

The 1994 INET paper “Internet Interaction Pinged and Mapped,” written with Carl-Mitchell and Gretchen Phillips, is a useful specimen. The team reported probing roughly 4,500 domains, 1,428 of them in Europe, six times each day. For the published European maps, it used data from eighteen Wednesdays between December 1993 and March 1994. The paper also said that probes were spaced to avoid unnecessary load, so the displayed time could differ from the actual probe by as much as two hours.

Those details change what the image can prove. Round-trip time from the MIDS observation point can show temporal and geographic patterns in the returned samples. It cannot, by itself, locate congestion, measure every user’s experience or establish that a provider owns the cause. Missing responses may reflect filtering, outage, name data, routing or the endpoint. A colour on a continent is evidence from one method and one location—not the continent’s intrinsic network condition.

The 1996 MIDS presentation on Latin America and the Caribbean made another disciplined choice: it did not ask one chart to answer every question. It separated Web servers, total hosts, host growth, hosts per capita and city-level distributions. It also asked registries and local participants for geographic information. A country-code domain can be administered or used away from the place named by its suffix; geography required evidence beyond the string.

Different measures can tell different, equally real stories. Total hosts may favour a large country. Hosts per capita can expose intensity. Growth over a year can reveal a small base accelerating. Web-server counts track one service rather than all connectivity. Latency maps show what a probe can reach and how long the exchange takes. None is “the Internet.” Together, with their definitions intact, they make the system more legible.

That legibility had economic force. The Internet Hall of Fame recounts that a 1994 Matrix Maps growth chart caught Jeff Bezos’s attention before Amazon. The anecdote is evidence about how measurement can affect perception, not proof that one graph caused a company. Investors, public agencies and network operators already had their own incentives and information. A visual time series could nevertheless turn distributed technical change into a claim that a decision-maker could compare with other opportunities.

The same force creates an incentive to overstate. A vendor benefits when a host count becomes a user count. A government may prefer national totals that follow ccTLD labels even when systems are elsewhere. A network may display average latency while hiding failed probes or inconvenient regions. Once a polished map circulates without its method, the uncertainty is separated from the conclusion while the conclusion keeps travelling.

The later history of the Internet Domain Survey shows why a measurement series must publish its own limits. ISC explains that the original survey walked domain data and counted addresses it could discover. Restrictions on DNS zone transfers reduced visibility. A later technique queried reverse DNS across delegated address space and counted a different object, making old and new results difficult to compare. Virtual hosting, aliases, missing reverse records and unreachable systems complicated the meaning of a “host.”

ISC eventually stopped the survey because an IPv4-only series had become misleading as IPv6 use and private networks grew. That decision is part of the evidence, not an admission that the historical work was worthless. A responsible measure can expire. Preserving the old method, the discontinuity and the reason for retirement makes the record more useful than silently extending a familiar curve beyond the world it can see.

Quarterman’s contribution is best understood in this discipline of visible incompleteness. He helped give the public maps of a network that did not have one centre, while his own writing kept returning to definitions of participation and reachability. The cartographer’s authority did not come from seeing everything. It came from showing what had been selected, from where, by which technique and for what question.

That lesson is especially important now that infrastructure dashboards can update continuously. Real-time display does not eliminate the old boundary problem. Telemetry still begins with deployed sensors, reachable interfaces, customer definitions, sampling rules and retained fields. A dashboard can be current and structurally blind at the same time.

The durable unit of trust is therefore not a dot, a curve or a global total. It is a measurement joined to its scope. Quarterman made the early Internet visible, but his sharper legacy is that visibility itself had to be engineered. Before drawing the network, the map had to say which Internet it meant.

Sources