Hunter's 1940 water-demand table, checked against his own 1% rule

We rebuilt Hunter's 1940 water-demand method, matched his printed table, found where it is looser than his own 1% rule, and shipped a free pipe sizer.

Every US plumbing code's water-supply fixture-unit table descends from one report: Roy B. Hunter, Methods of Estimating Loads in Plumbing Systems, National Bureau of Standards BMS65, December 1940 (NIST scan). His idea fits in two lines. A fixture that draws q gpm for t seconds once every T seconds is running with probability p = t/T at a random instant. With n such fixtures, the number running is binomial. Size the pipe for the number m that is exceeded no more than 1% of the time.

Today's run rebuilt that model from the paper, in a new free tool: water supply pipe sizing. This is what checking it against the original turned up.

His table reproduces

Hunter's Table 1 gives the busy-period inputs: flush valves 9 s in 300 s at 27 gpm, flush tanks 60 s in 300 s at 4 gpm, bathtubs 60 s in 900 s at 8 gpm. His Table 2 lists, for each design number r, the group size n where he steps up to it and the probability of exactly r running there. Our probability function, computed in logarithms, matches his printed value to the fourth decimal in 13 of the 14 legible rows. The odd one out is 31 bathtubs: we get 0.0115, the paper prints 0.0118. That is a hand-calculation slip on one side or the other, or a misprint. (The flush-tank column is garbled in the scan's text layer, so we left it out.) At every tabulated n, the exact rule gives his m.

But the exact rule steps up sooner

Hunter's text sets the criterion as the count that is "not exceeded more than 1 percent of the time". That is a tail: P(more than m running) ≤ 1%. His table steps up when P(exactly r running) reaches 1%. He justified the substitution with a series bound: the terms past m shrink at least by half each step. The bound is close for small groups and drifts for large ones:

Flush valves, design numberHunter's table steps up atExact 1% tail rule needs it from
43029
66662
810799

At 106 flush valves the table still says 7. The chance that more than 7 are running is 1.47%, not 1%. Bathtubs show the same pattern (9 tubs from 55, not 59). It is one fixture's flow at most. Next to how rough "once every five minutes" is as a model of a 1930s office, it does not matter. Still, it is a small, checkable gap between the method as stated and the method as printed, and we had not seen it noted. The tool follows the stated rule, and a test pins the whole comparison table to the engine.

Mixing fixtures without fixture units

Hunter handled a mix of fixture types by converting each to "fixture units", weights read off his curves, which is where the code tables come from. A computer does not need the weights. Each type's flow is q times a binomial count, the types are independent, so the total flow's distribution is just their convolution. The tool computes it on a 0.1 gpm grid and reads off the flow exceeded 1% of the time (or whatever share you choose). A brute-force enumeration test checks the convolution.

The difference is large. In the tool's example two-bath house, with illustrative timings (they are assumptions, not measurements), the service pipe works out as follows:

  • every fixture open at once: 31.3 gpm;
  • each type sized on its own by Hunter's single-type rule, then added: 25.3 gpm;
  • the mix worked as one distribution: 10.8 gpm.

Adding per-type peaks assumes the rare busy moment for showers is also the rare busy moment for toilets and the kitchen. That is what makes the sum of separately sized peaks so pessimistic.

Then the pipe

The demand feeds a pipe tree. Each segment carries the fixtures it feeds and everything beyond. The tightest path's pressure budget sets one allowable friction rate: static pressure, less meter and backflow losses, the minimum at the fixture and 0.433 psi per foot of rise, divided by developed length. In the example that is 15.22 psi per 100 ft. Each segment then gets the smallest copper or PEX size within that rate and the velocity limit (Hazen-Williams; inside diameters from the ASTM B88 outside diameter and wall), and the tool walks every path to report the pressure left at each fixture.

There is one more step, which code tables cannot take because they never see the pressures. The uniform friction rate comes from the worst path, so every shorter path has pressure to spare. With Trim switched on, the tool steps segments down a size, longest first, while every end still holds the minimum. In the example, both upstairs bath branches drop from 3/4 in to 1/2 in, and the hall bath still gets 24.2 psi against a 20 psi minimum.

What it is not

It is not your code's fixture-unit table, and an inspector may require that table. Fixtures are treated as independent, Hazen-Williams is a cold-water correlation, and fittings are a percentage allowance. The disclaimer on every page says so. What it adds is the model behind the table, worked exactly for your fixtures. You can see why a pipe is the size it is, and how much pressure is left at the far shower.

Next, if people use it: a hot-water branch with a recirculation loop, and a side-by-side with a published modern probabilistic method for houses, where Hunter's congested-building timings are known to oversize.

All writing · More in Tools and projects · All tools and research