Summary

  • On 1 February 1996, Olivier MJ Crepin-Leblond at Imperial College London ran a traceroute to www.mg.refer.org, which resolved to a host named syfed-tana.mg.refer.org in Antananarivo.
  • The destination answered one probe after a long international path; two probes timed out. This proves a specific moment of reachability, not routine performance or nationwide Internet access.

The archive preserves a small but unusually concrete moment in Madagascar’s Internet history. At 14:53 UTC on 1 February 1996, Olivier MJ Crepin-Leblond sent a traceroute from the UK to the French-language site www.mg.refer.org. The name resolved to syfed-tana.mg.refer.org, address 193.104.63.18. The command allowed up to 30 hops and used 40-byte packets. The surviving output records a route, not a survey: one operator, one starting point, one destination and one run.

Its path begins inside Imperial College, passes through JANET and the LINX exchange, then through EU.net nodes in Canterbury, Amsterdam and Paris. The trace next names isdn-gw.aupelf.fr, an AUPELF gateway, before reaching an address in the Madagascar network. Hop 14 returns no response. At hop 15, the destination responds once in 9,383.95 milliseconds; the other two probes time out. Crepin-Leblond closes with a short conclusion: the website was already up and running.

That last line is easy to overread. The result establishes that at least one probe from this UK host reached the named web server at that moment. It does not establish that the page loaded reliably for other users, that the server stayed online, or that residents could obtain Internet access. Nor is 9.38 seconds a representative latency figure. It is one reported response among three attempts, and the archive offers no repeat test or explanation for the delay.

Earlier records place the probe in a staged institutional setting. A 7 November 1995 form describes Centre SYFED/REFER of AUPELF-UREF in Antananarivo as a secondary node serving scientific and university users. It lists email and FTP as available services; Gopher, Archie, Netscape and Mosaic were expected soon. Access from abroad was described through a modem and X.25, while a web server was still in the future. On 22 December, another message said that a French-language “Madagascar Contact” site was available at the same address, though still under construction. The accounts capture different dates and service descriptions.

They do not amount to a continuous uptime record.

The traceroute is useful because it turns an abstract claim—an institution in Madagascar could be reached from abroad—into a dated operational observation. It also displays the limits of that observation. Traceroute lists devices that return messages; asterisks do not necessarily mean that traffic stopped, because routers may forward packets without answering diagnostic probes. The printed hops are not a complete physical cable map, and the visible route cannot establish the reverse path.

In network histories, “connected” can describe several different conditions: a route exists, a server answers, a service is usable, an institution has access, or the public can afford a connection. The 1996 record supports the first two only in a narrow, momentary sense. The November and December notes add evidence of services aimed at academic users and an emerging website, but they do not show how many people used them or what access cost.

The safest reading is also the most revealing. Madagascar’s web host was reachable from a British university through a chain of research and commercial networks, and the test itself encountered silence and a strikingly slow final reply. That is neither a national launch date nor a performance benchmark. It is a glimpse of international reachability becoming observable while the broader conditions of reliable, widely available access remained outside the measurement.

Sources