Summary

  • On 28 April 2001, the Bergen Linux User Group tested an implementation of RFC 1149 using pigeons as physical carriers between two computer endpoints.
  • Its log shows nine pings sent, four returned and 55% packet loss; the four measured round trips averaged about 87 minutes.
  • The test demonstrated a narrow end-to-end path through a manual physical adapter. It did not establish dependable access, commercial service or standards-track interoperability.

A protocol had to leave the page

RFC 1149, published on 1 April 1990, described how to carry IP datagrams on avian carriers. It called the method experimental and not recommended; its opening described high delay and low throughput. Eleven years later, the Bergen Linux User Group (BLUG) and local pigeon fanciers set out to make a packet cross a real route. BLUG’s event page says the test took place in Bergen on 28 April 2001. Its meeting notes show that the group had planned a public demonstration, assigned separate work on the bird-and-packet handling and on the computer interface, and expected weather and training to matter.

The surviving terminal log makes the system unusually legible. One computer exposed a point-to-point interface named tun0, with a peer address of 10.0.3.1; the operator then ran ping across it. The interface reported a 150-byte MTU. The documents do not give a complete software or radio/network diagram, so the safe description is the one the record supports: the computers exchanged ICMP probes through a software interface and a physical pigeon-carried packet, with the remote packet scanned and checked before it was accepted.

That manual boundary is the story. IP did not ride inside a bird as a continuous electrical signal. An adapter had to turn network data into a carried object, get it to the other endpoint, recover the data and hand it back to software. The bird was the link; people and scanners were part of the system.

The result was measurable—and limited

BLUG’s event page lists noon as the event time. The later write-up says the ping began at approximately 12:15, while the preserved shell transcript starts at 11:24 and ends at 14:14. Those timestamps describe different parts of the session and do not support a cleaner single start time. The narrative says the birds initially joined a neighbour’s flock; after roughly an hour, some broke away toward the destination. The first returned packet was scanned, its optical-character-recognition errors were checked by hand, and the reported ICMP reply arrived.

The transcript records nine probes and four replies: 55% packet loss. Their round-trip times ranged from 3,211,900.8 to 6,388,671.9 milliseconds; the log’s average is 5,222,806.6 milliseconds, about 87 minutes. The write-up says six return pigeons arrived, two without packets, while the log’s four replies account for the packets that completed the return path. These are results from one demonstration, not a measure of a repeatable service or of pigeon transport generally.

The group declared the implementation a success and hoped for other implementations so it could test interoperability. That wording matters. It was a successful demonstration in the group’s stated sense: a packet traversed the constructed path and software accepted the reply. The records do not show a second implementation, a standards-track upgrade or an operational service. RFC 1149 remained explicitly experimental; RFC 2549 later extended the joke with a quality-of-service memo, not a service guarantee.

The episode is useful because it separates three claims often blurred in technology history: a specification can be implemented; a packet can make a round trip; and a network can be dependable enough for users. Bergen demonstrated the first two under controlled conditions. Its own packet loss and hour-scale delay make the third a different question.

Sources