Field notes on things that run themselves
Perfect Efficiency Is an Infinite Line
The line at a checkout can hold three people all afternoon. Not the same three: the woman with the flowers pays and is gone, someone with a basket joins, and the line itself stays three people long from lunch to closing. By evening a hundred strangers have passed through it. No one who joined consulted anyone who left; the cashier did not choose it; the length is written nowhere. Of all the standing patterns in this series — flames, glaciers, clouds of mud — this is the one you can walk into, hold up with your own body, and leave.
Two numbers make it. Customers arrive at some rate — say ten an hour. The cashier clears them at another — say twelve, when there is anyone to clear. Notice the kind of balance. No. 84’s chemostat held its level because an operator bolted a pump to a tank and picked the rate. Here nobody picked either stream: shoppers arrive by private whim, the cashier scans as fast as scanning goes, and the length of the line is a negotiation between two processes that never met — a number that hangs in the air and belongs to no one.
Be careful what is actually standing. The length never holds still — five now, one in a minute, empty and rebuilt by two o’clock. What stands is the odds: over an hour the line spends a fixed fraction of its time empty, a fixed fraction three deep, a fixed fraction seven; Thursday the fractions repeat. In the textbook’s simplest model — arrivals random, service times random, one server — the odds take a clean shape: each extra person of depth is a fixed factor less likely than the last, all the way up. The members pass through; the probabilities do not move.
Note what the line is not. No. 9’s phantom jam is a wave — congestion rolling backward through traffic, different cars each second, going somewhere. A queue is a stock: pinned to its server, fed at one end, drained at the other, going nowhere at all.
The arithmetic of its size is insultingly small. Take the fraction of time the server is busy — utilization, arrivals divided by service. The fraction of time the line stands empty is exactly the rest. And the average number present, counting the one being served, is the first divided by the second: busy time over idle time. That identity is the whole story. Busy half the time: a line of one. Eighty per cent: four. Ninety: nine. Ninety-five: nineteen. Ninety-nine: ninety-nine. The traffic barely grew across that list; what was eaten was the denominator. At always-busy, in the standard model, no steady length exists — the line wanders upward without a ceiling. Perfect efficiency is not a fast line. It is an infinite one.
The textbook’s own example: ten customers an hour. A server handling eleven an hour carries an average line of ten; a server handling twelve — nine per cent faster — carries five. The lever works in reverse with the same violence — why systems tuned to ninety-odd per cent busy feel broken to everyone in them.
But the cliff is not made of load. It is made of luck. Run the same nine-tenths utilization as clockwork — an arrival every ten minutes, each served in nine — and the queue is zero, forever. The line exists because randomness clumps: three arrivals in a minute, one order that takes an age. The store cannot bank against the clumps: an idle second cannot be saved, while a burst must be paid in full, one customer at a time. Chance deposits; idleness expires. Steady the service to clockwork and the waiting in line halves, exactly. A queue is variability made visible — a standing inventory of coincidences with nowhere else to go.
One law under all this needs none of the model’s assumptions: the number standing equals the rate through times the time inside. Philip Morse published it in 1958 and challenged readers to find a queue that broke it; John Little proved in 1961 that none exists — any arrivals, any service, any discipline, any number of servers. Bookkeeping, indifferent as a conservation law. This series has used it for years without saying so: No. 67’s rod outer segment holds about a thousand discs and builds eighty or ninety a day — divide, and a disc stands a week and a half before the tip eats it. Your retina is a checkout in membrane; the wait is why it is always ten days old.
The science began at this register. In 1908 A. K. Erlang joined the Copenhagen Telephone Company to work out how many circuits a city needs; his first paper, in 1909, showed random calls obey Poisson’s law — the randomness the model still runs on. At the centre of what his successors built sits the finding a century of managers has hated: the idle fraction is not the system failing. It is what the shortness of the line is made of — busy over idle. Spend the idle down to nothing and the difference does not vanish; it goes into storage, and the storage is people, standing. A queue is randomness given a place to stand. What keeps the place small is the rest you did not optimize away.
One loop I’m watching
Next: your blood replaces about two hundred billion red cells a day — a couple of million every second — and the count barely moves for decades. The level is held by a sensor that never counts cells at all: tissue in the kidneys reads the oxygen instead, and dials a hormone up or down until supply matches the wear. This issue’s arithmetic reappears wearing marrow — stock equals rate times lifespan — but with the controller the checkout never had. Next time.
Tip: the ← and → arrow keys move between issues.
Prefer feeds? Follow by RSS or JSON Feed.
Want something playable between issues? Thought Toys has its own email list and RSS feed.
New to The Standing Wave? Start here →