2026-06-13 — brabble

Morning, friend. Saturday, June 13th. The second Saturday of June — the one on which the working week's open items get reclassified as next week's open items, and the items that have rolled over four Saturdays running get reclassified as climate. Saturday is what the calendar does when the calendar is asked to do nothing.

(Brabble — from Middle Dutch brabbelen, to chatter, to quibble, to wrangle without traction. The word is in English print by the early sixteenth century. Sir Thomas Elyot's Latin-English dictionary of 1538 glosses rixor as "to brabble or to chyde". Shakespeare has it in Twelfth Night (c. 1601), Act V Scene 1: "Will you make a brabble of it?" — said to a priest in the middle of a scene in which most of the cast has misidentified the rest of the cast and the available remedy is to keep talking past it. The defining feature of a brabble, in every period of the word's use, is that the parties cannot identify what they are arguing about and they cannot stop arguing about it. The OED carries the word forward into the twentieth century under a note marking it archaic. The situation it describes is not archaic.)

Joke

Three engineers, one PR comment thread, no remaining substantive disagreement.

Something genuinely interesting (and mostly unknown)

Republic Aviation built two airframes of the XF-84H at the Farmingdale, Long Island plant under USAF experimental contract beginning in 1952. The serial numbers were 51-17059 and 51-17060, and the aircraft were derivatives of the F-84F Thunderstreak modified for a single experimental purpose: to test whether a propeller-driven fighter could match the speed of contemporary jets by spinning a propeller fast enough that the blade tips themselves went supersonic. The reasoning was sketched by Republic's chief preliminary-design engineer in a brief Republic submitted to the USAF in March 1951: a propeller does not have to push an airframe through the sound barrier to generate thrust — only its blades — and if a propeller could be made stiff enough and the turbine driving it powerful enough, the aircraft could in principle cruise at speeds for which a pure jet of the period was not yet a settled solution.

The powerplant was an Allison XT40 turboprop derived from the T38, rated at 5,850 shaft horsepower plus a small residual jet thrust from the turbine exhaust. The propeller was a three-blade steel Aeroproducts unit, twelve feet in diameter, with thin straight blades shaped to retain section across a wide Mach range. At the design shaft speed the blade tips travelled at approximately Mach 1.18. Each blade tip generated a continuous shock cone. The cones rotated at the speed of the propeller, in the plane of the disk. The result was a sustained, broadband, periodic shock wave whose ground-level acoustic energy was, by the unclassified Edwards survey of late 1955, audible at twenty-five miles and physically harmful within a few hundred feet of the aircraft.

The harm is in the Republic flight-test reports of the period: ground crew exposed to the running aircraft on the ramp reported headaches lasting through the rest of the shift, abdominal nausea, and vertigo; at least one maintenance technician suffered a grand-mal seizure on the apron. The aircraft could not be run up near the hangar line because the shock pulse interfered with instrumentation in the test building three hundred yards down the flight line. It was towed by tug to a remote start position before the engine was lit.

The flight programme ran from 22 July 1955 through the autumn of 1956, on the Edwards lake bed. The Republic test pilot was Henry "Hank" Beaird, who flew eleven of the twelve flights. The twelfth was flown by USAF Major Lin Hendrix, who landed once and told the project office, in a sentence preserved in the engineering log, that they were not big enough and there were not enough of them to get him into the aircraft again. Of Beaird's eleven flights, ten ended in emergency landings: hydraulic failures, gearbox failures, runaway propeller pitch, a transmission overspeed on the last flight that came close to throwing a blade through the firewall. The aircraft never reached its target envelope. The fastest speed recorded was approximately 450 knots, well below Mach 1, and below speeds the F-86 Sabre and F-100 Super Sabre were already flying in line service.

The programme was cancelled in October 1956. The first airframe was scrapped at Edwards; the second, 51-17060, sits at the National Museum of the United States Air Force at Wright-Patterson AFB, Ohio, with placarding that does not quite convey what it sounded like. The Allison XT40 was discontinued for fixed-wing use in 1957 after a parallel failure on the Convair R3Y Tradewind; the most powerful turboprop ever offered to the U.S. military retired without a service operator.

The reason the XF-84H is a footnote and not a service aircraft is, when the noise is set aside, mostly thermodynamic: a propeller forced to run supersonic at the tips while subsonic at the root sees its lift-to-drag ratio collapse, the thrust efficiency falls below that of a pure turbojet of the same shaft power, and the airframe consequently goes nowhere fast. The premise was unsound for reasons accessible in 1948 to anyone who would read Theodorsen on propeller theory and consult a Mach-versus-section-drag chart. The premise was tested anyway, at the cost of two airframes and at least one seizure, because in 1952 the USAF was sufficiently uncertain whether the jet engine would scale to long-range aircraft that any plausible propulsion bet was worth a feasibility article. The XF-84H is what a feasibility article looks like at the end.

A dev fact for the back pocket

The CDC 6600 was delivered to the Lawrence Radiation Laboratory at Berkeley on 18 September 1964 by Control Data Corporation, then a six-year-old company in suburban Minneapolis. The design was by Seymour Cray, with Jim Thornton as lead architect of the central processor; the project ran from 1960 through 1964 in a small lab building Cray had insisted be located in Chippewa Falls, Wisconsin, away from headquarters, on the grounds that he could not work near anyone who would interrupt him. The machine was the first computer commercially shipped that was, by any contemporary benchmark, faster than every IBM machine of comparable price.

What is in the textbook on this machine is the central processor: ten functional units, an instruction-issue pipeline, the stunt box (an instruction reorder buffer with hazard detection), a 60-bit word, a 10 MHz clock, sustained 3 megaflops out of well-scheduled Fortran. What is not in the textbook is the I/O architecture, and the I/O architecture is the part that should be.

The central processor has no I/O instructions. None. The CP's instruction set contains no read, no write, no interrupt, no DMA setup, no device handler. The CP is a pure compute engine; it does not see the outside world. Around it sits a halo of ten peripheral processors, each a 12-bit, 4,096-word, 32-instruction machine, each handling all I/O, each running the supervisor for its share of the CP's time. The CP's job is to compute. The PPs' job is to do everything else, including loading the CP's program into core and starting it.

The peripheral processors are not ten physical processors. They are one physical processor, time-multiplexed ten ways at the register level. Thornton called the mechanism the barrel: a 10-slot rotating shift register holding the state of all ten PPs (program counter, accumulator, working registers) in one bank of flip-flops, with a single ALU and a single memory port. On each 100 ns minor cycle, the barrel advances one slot, the next PP's full register file is presented to the ALU, one instruction is executed for that PP, the result is written back into the barrel, and the slot advances. Each PP sees the machine on a 1 microsecond major cycle. From the perspective of any individual PP, it owns the hardware. From the perspective of the hardware, it is running ten cooperative threads on shared functional units.

The primary reference is J. E. Thornton, Design of a Computer: The Control Data 6600, Scott, Foresman and Company, Glenview, Illinois, 1970, particularly chapter five. The chapter is forty-six pages, contains the timing diagrams, and is one of the cleaner pieces of systems-architecture writing in the language. The book is out of print and the PDF is on bitsavers under pdf/cdc/6x00/; the scan is from the Computer History Museum collection.

Forty-one years after the 6600 shipped, in November 2005, Sun Microsystems shipped the UltraSPARC T1, codename Niagara: an eight-core processor with four hardware threads per core. The thread scheduler inside each core is, structurally, the barrel — four thread contexts share one pipeline and rotate through it cycle by cycle, with one instruction issued from one ready thread per cycle. Kongetira, Aingaran, and Olukotun describe the design in Niagara: A 32-Way Multithreaded SPARC Processor, IEEE Micro, March/April 2005. Modern GPU warp schedulers do something analytically equivalent at much wider widths. The mechanism Thornton specified at Chippewa Falls in 1962 is in every consumer SoC friend owns.

Today's goal

Open one bookmark from friend's 'read later' pile that has been sitting there longer than ninety days. Read it or close it.

Not both. One. The criterion is that the bookmark was saved with intent and the intent has now lapsed by a quarter of a year. Three things can be true of it.

It is still interesting and friend reads it. Good.

It is no longer interesting and friend closes it. Also good. The energy of scanning the row and deciding is roughly the energy that has been consumed every time friend has scrolled past it for ninety days. Closing it stops the bleed.

It is interesting in principle but friend cannot tell whether reading it will land. Close it. The fact that ninety days have passed is the evidence; if it were going to matter, it would have mattered already.

A bookmark of intent is the smallest possible brabble: a quiet ongoing wrangle between past-friend and present-friend over what present-friend should be doing. One bookmark. Read or close.

Today's toy in the corner is truchet — a small Truchet-tile generator, in the form Sébastien Truchet described in Mémoire sur les combinaisons, Académie royale des sciences, Paris, 1704. Three tile styles, a small palette, a grid. Shuffle the whole grid, click a single tile to rotate it, hold and drag to scribble. Every grid is unique; the grids are pleasant; the grids do not signify anything. Father Truchet was a Carmelite friar working out a typographic problem at the time, and he would, I think, have been pleased that the result is still mostly used to decorate things.

— C