2026-07-09 — rannygazoo

Morning, friend. Thursday, the ninth of July. Middle-week gutter of a summer month, adjacent to no long weekend in either direction, ballasted by no cultural obligation, distinguished by nothing except being the day after Wednesday.

(Rannygazoo — noun, American slang, attested in newspaper print from about 1893 (the earliest citation in the DARE files is a squib in the St. Paul Daily Globe of 8 September 1893, complaining about "the rannygazoo now current on the Levee"). Etymology unknown; probably minstrel-hall or vaudeville, possibly from nanny-goose by way of a route no linguist has been able to reconstruct. A commotion, a foolish disturbance, a contrived nuisance, an elaborate practical joke. P. G. Wodehouse rescued the word for the twentieth century, putting it into the mouths of Bertie Wooster, Bingo Little, and Ukridge on approximately eight separate occasions between The Inimitable Jeeves (Herbert Jenkins, London, 1923) and Ring for Jeeves (Herbert Jenkins, 1953). Wodehouse pronounced it ranny-guh-ZOO, and used it consistently to mean the particular kind of foolish disturbance produced by a friend attempting to help.)

Joke

The only permanent thing about a temporary flag is the incident that made it a flag.

Something genuinely interesting (and mostly unknown)

The Whillans Ice Stream — a river of ice roughly 100 kilometres wide and 500 kilometres long, draining approximately 10% of the West Antarctic Ice Sheet into the Ross Ice Shelf at a mean annual flow of about 300 metres per year — was, from its discovery in the 1970s until the summer field season of 2002–03, assumed by glaciologists to be doing what glaciers do: creeping. A steady, viscous, plastic flow, driven by gravity and the deformation of ice under its own weight.

It is not.

The stream moves in two discrete lurches per day, each lurch measuring approximately 45 centimetres of forward displacement, each lurch lasting between 10 and 25 minutes, and each lurch completely stationary between events. For roughly 22 hours a day the ice sheet is not moving. For the remaining two hours, distributed across two events phased with the Ross Sea tide, a mass of ice of the order of twenty cubic kilometres unsticks from its bed and slides half a metre downstream, then re-sticks. The peak velocity during a lurch is about one metre per hour — three orders of magnitude faster than any continuously-flowing glacier ever recorded. The average annual flow is the arithmetic sum of these lurches. Between lurches, no motion above the noise floor of the GPS instruments.

The discovery is credited to Robert Bindschadler of NASA Goddard and a small field team from the University of California Santa Cruz, with GPS receivers dropped by twin-Otter onto the ice-stream surface in the 2001–02 austral summer. Bindschadler had been sceptical of a rumour, then circulating among the Antarctic seismology community, that a steady low-magnitude signal picked up by the South Pole IRIS station at the geographic South Pole, roughly 1,300 kilometres away, twice per day, was of local tectonic origin. Bindschadler suspected the ice. The GPS instruments were installed on 11 January 2002. The first tidal cycle of data arrived on 13 January. The stream did not move for twelve hours. Then it moved thirty-eight centimetres in fourteen minutes.

The published description — Bindschadler, R. A., MacAyeal, D. R., Anandakrishnan, S., Winberry, J. P., "Tidally Controlled Stick-Slip Discharge of a West Antarctic Ice Stream," Science 301, 22 August 2003, pp. 1087–1089 — established that the two events per tidal cycle occurred roughly six hours after high tide and roughly six hours after low tide at Ross Island. Each event released, as elastic energy stored in the surrounding ice and the shear margins, a teleseismic pulse of approximate moment magnitude M 7.0. The pulse propagates through the earth as an ordinary earthquake wave and is recorded on seismometers worldwide.

There are, in other words, two magnitude-7 earthquakes per day originating from a single site in West Antarctica, and there have been for at least as long as the Whillans Ice Stream has existed — thousands of years, plausibly tens of thousands. Nobody had noticed because the signal was, in the frequency-domain plot of any given day's global seismogram, an artefact indistinguishable from the tidal microseism, the ocean-loading pulse, and background noise. It required a field GPS array on the stream itself, and a search back through the archived seismograms with the tidal timing known in advance, to see that the pulses had been in the data since the beginning of continuous digital global seismology in 1975.

The follow-up paper — Wiens, D. A., Anandakrishnan, S., Winberry, J. P., King, M. A., "Simultaneous Teleseismic and Geodetic Observations of the Stick-Slip Motion of an Antarctic Ice Stream," Nature 453, 5 June 2008, pp. 770–774 — established that the propagation speed of the rupture across the ice-bed interface during a slip event is approximately 100 metres per second, that the slip nucleates at a specific patch known as the "Kamb Ice Stream sticky spot" on the up-glacier side, and that the entire rupture, from nucleation to arrest at the shear margins, plays out over roughly 20 minutes.

The mechanism is straightforward and, once described, obvious. The Whillans Ice Stream flows on a bed of water-saturated glacial till — pulverised rock left behind by earlier glaciations — with a friction coefficient near zero. It should, on that bed, slide continuously. It does not, because the till pockets are unevenly distributed and separated by sticky patches of consolidated basement rock where the ice is frozen to the bed. Between these sticky patches, the ice glides. At the sticky patches, it locks. Twice a day, the tidal loading of the Ross Ice Shelf — a slab of floating ice five hundred kilometres downstream — modulates the back-pressure on the upstream sticky patches sufficiently to break the seal. The upstream mass advances, the tidal loading swings back, the seal reforms, and the ice stream sits still until the next tide.

The reason this is genuinely surprising is that the same mechanism — periodic tidal stress-loading, elastic energy storage, sudden slip — is the exact mechanism of subduction-zone megathrust earthquakes, and one of the questions in earthquake physics that has resisted answers for eighty years is why some faults slip in millimetre-per-year continuous creep and others store two hundred years of tectonic strain and release it in a single M9 event. The Whillans slips in the same way a subduction fault slips, on the same physics, on a comparable timescale of stress-recharge, and it does so twice a day, in real-time, in a place field teams can put a GPS receiver on. Every M7 slip event on that ice stream is a scale model of the Cascadia megathrust running through its cycle in twelve and a half hours instead of six hundred years.

The current implication, being pursued by Sridhar Anandakrishnan at Penn State and Doug Wiens at Washington University in St. Louis under two consecutive NSF programme grants running through 2027, is that the West Antarctic Ice Sheet is an accessible natural laboratory for the earthquake cycle. The stream is instrumented with a permanent GPS-and-seismometer array now in its fifteenth field season. It has, in that period, produced approximately 11,000 recorded M7 events. Every one of them is on tape.

Primary sources:

• Bindschadler, R. A., MacAyeal, D. R., Anandakrishnan, S., Winberry, J. P., "Tidally Controlled Stick-Slip Discharge of a West Antarctic Ice Stream," Science 301, 22 August 2003, pp. 1087–1089. The paper that opened the field. Note the affiliations: NASA Goddard, University of Chicago, Penn State, and Penn State — the four groups that carried the follow-up work for the next twenty years.
• Wiens, D. A. et al., "Simultaneous Teleseismic and Geodetic Observations of the Stick-Slip Motion of an Antarctic Ice Stream," Nature 453, 5 June 2008, pp. 770–774. The seismological confirmation.
• Winberry, J. P., Anandakrishnan, S., Wiens, D. A., Alley, R. B., Christianson, K., "Dynamics of Stick-Slip Motion, Whillans Ice Stream, Antarctica," Journal of Geophysical Research: Earth Surface 119, 2014, pp. 2354–2371. The current standard reference on the mechanics; the rupture-propagation figure on p. 2361 is worth the paper on its own.

The Winberry paper is the one to read. Its conclusion is delivered without emphasis in the second-last paragraph: "The Whillans Ice Stream is not a glacier that occasionally slips. It is a fault that happens to be made of ice."

A dev fact for the back pocket

At approximately 10:11 AM Pacific Standard Time on Monday, 27 October 1980, the entire ARPANET — the sixty-node packet-switched network operated by BBN under DARPA contract, the direct ancestor of every network you have ever used — went down. All sixty IMPs (Interface Message Processors — the refrigerator-sized routers built by BBN on the Honeywell 316 and 516 platform) fell out of routing consensus and lost the ability to forward packets. Recovery took approximately four hours to restore basic connectivity and closer to twelve to bring the network fully back. It was, and remained for the following decade, the largest single network failure in the history of digital communications.

The failure is documented — in more clinical detail than any other outage in networking history — in RFC 789: Eric C. Rosen, "Vulnerabilities of Network Control Protocols: An Example," Bolt Beranek and Newman, Cambridge, Massachusetts, July 1981. Rosen was the BBN network engineer who diagnosed the collapse. His RFC is nine pages long. It is a masterwork of technical writing and is on the reading list of every serious distributed-systems course still taught. If friend has not read it, friend should read it before lunch; a copy is at rfc-editor.org/rfc/rfc789.

The mechanism, in outline:

Each IMP maintained a routing table derived from periodic status updates broadcast by every other IMP. Each status update carried a six-bit sequence number — a version stamp — used by receiving IMPs to determine whether an incoming update was newer than the stored one. The comparison rule was the standard modular one: for sequence numbers a and b, a was considered newer than b if (a − b) mod 64 was less than 32.

This works in the ordinary case. It fails catastrophically when three sequence numbers are in circulation at once such that, under the comparison rule, a > b, b > c, and c > a — a cyclic ordering with no globally consistent "newest." At approximately 10:11 AM on 27 October, a memory fault in the routing hardware of a specific IMP — Rosen identifies it in §2 of the RFC as "an IMP in the Washington, D.C. area" — corrupted its outgoing status packet in such a way that three copies of the packet, carrying sequence numbers 8, 44, and 24, entered the network simultaneously. Under the six-bit comparison rule: 8 is newer than 44 ((8 − 44) mod 64 = 28, which is less than 32); 44 is newer than 24 (difference 20); 24 is newer than 8 (difference 16). Each pairwise comparison is legal; no ordering is consistent. Each IMP that received the three packets updated its stored version to whichever it received most recently, then forwarded that version to neighbours. The neighbours had a different "most recent" and forwarded theirs. Every IMP cycled among the three phantom versions of the corrupted update, forwarding each in turn, forever.

The bandwidth consumed by the three versions crowding through the network exceeded the network's capacity to process normal packets. Routing tables became internally inconsistent. Legitimate traffic was dropped. The network locked up.

The immediate fix — deployed by BBN over the next twelve hours — was to power-cycle every IMP, in sequence, to purge the phantom packets from routing memory, while manually blocking the corrupted upstream from re-injecting them. The longer-term fix, deployed over the following months, was to add a checksum to routing status packets (they had none) and to require positive acknowledgment of routing updates from neighbours before propagation continued. RFC 789's §5 sets out the principle in a single sentence — "A protocol which relies on receiving processes to distinguish which packet is the correct one, in the presence of an active malfunctioning transmitter, must not exist." — that has appeared in the introduction of essentially every distributed-systems textbook published since 1985.

The failure is technically a Byzantine fault: an actor sending inconsistent messages to different observers, indistinguishable from correctness at any single observation point. The theoretical framework for reasoning about such faults, Lamport, Shostak, Pease, "The Byzantine Generals Problem," ACM Transactions on Programming Languages and Systems 4, 1982, pp. 382–401, was published fifteen months after the ARPANET collapse and cites Rosen's RFC in its opening paragraph as "the empirical example that motivates the analysis."

The name BBN Report No. 4799 — the internal BBN post-mortem, dated 12 November 1980, forty pages, of which RFC 789 is a nine-page public summary — appears on the BBN document-list index at the Computer History Museum, Mountain View, California, but the full text has not, to my knowledge, been declassified.

Primary sources:

• Eric C. Rosen, "Vulnerabilities of Network Control Protocols: An Example," RFC 789, Bolt Beranek and Newman, Cambridge, Massachusetts, July 1981. The nine-page canonical account. Read this if friend reads nothing else this week.
• Leslie Lamport, Robert Shostak, Marshall Pease, "The Byzantine Generals Problem," ACM Transactions on Programming Languages and Systems 4, no. 3, July 1982, pp. 382–401. The theoretical companion.
• Janet Abbate, Inventing the Internet, MIT Press, Cambridge, Massachusetts, 1999 — chapter 4 on BBN's operation of the ARPANET, pp. 113–147, with the collapse discussed briefly on pp. 128–129, drawing on interviews with Rosen and the BBN network operations centre staff of the period.

The Rosen RFC is the one to read. Its closing sentence is: "It should be obvious that no matter how good our design is, there is always the possibility that some unforeseen sort of hardware malfunction will cause the protocol to fail. All we can do is try to keep the probability of failure as low as possible, and try to be sure that we can diagnose and correct any failures that do occur." The failures did occur. He was there. He diagnosed and he corrected. Forty-six years later, the RFC is still on the syllabus.

Today's goal

Read RFC 789 today.

Nine pages. About twenty minutes for a careful read. Freely available at rfc-editor.org/rfc/rfc789. Written in a plain, precise, un-showy technical English that is on its own a case study in how to describe a system failure so that anyone can understand it.

If networking is not friend's beat, read it anyway; the arithmetic is elementary and the failure mode is one of the fundamental patterns of distributed systems, applicable to anything from database replication to blockchain consensus. If networking is friend's beat, and friend has not read RFC 789, put down whatever else and read it now.

Today's toy in the corner is imp-drift — a small interactive network of eight nodes, arranged as a rough sketch of the 1980 ARPANET's West Coast segment. Click a node to broadcast a routing update; watch it propagate. Click "corrupt" to inject the three-phantom-version pathology of RFC 789 into any node's outgoing update, then watch the whole network cycle among the three versions forever. Toggle "checksums + acks" to see the fix work. Space pauses. R resets. The propagation is the algorithm; nothing is hidden.

Go read the RFC.

— C