The collection

Every essay — page 6

Page 6 of 13, continuing through the fields in the same order.

What a bound is Counting The floors What the machine does Structures Two parameters The other axis When the algorithm flips a coin What the libraries do When it does not fit One pass, and no room The data that is not a number When the algorithm is a table The index that replaces the text What is taught wrongly Ladders Objects Search

What the machine does

The operation count is not the running time. Locality, cache lines and branch behaviour decide the constant, and they rank algorithms differently from the textbook count.

Thompson's NFA, a set of states43.724 to buildthe subset DFA, built in full1.011,297 to buildthe subset DFA, built on demand41.824 to buildthe state set in machine words37.224 to buildoperations per character; the note is what was paid once, before the first character4,096 characters · k = 643.7 down to 1

What a character costs on four machines

Forty-four operations, one, forty-two, and a number that moves. Four machines for one language, with the construction charged separately from the steps, because a machine that is free per character paid eleven thousand operations before the first one.

8 figures
rounds, if every ready cell ran at oncerow by row65,793column by column65,793anti-diagonal by anti-diagonal513reads that miss the cacherow by row6.3%column by column28.2%anti-diagonal by anti-diagonal31.1%fully associative · 32 lines × 8 elements · LRU263,169 reads and writes per order

The order with the best depth

An edit-distance table can be filled row by row, column by column, or one anti-diagonal at a time, and the anti-diagonal order is the one that needs the fewest rounds — 513 against 65,793 on two strings of 256 characters, because every cell on an anti-diagonal is independent of the others. Stored the usual way, row by row, it also misses the cache on 31.1% of its reads, where row order misses 6.3%. The order that is best for parallel work is worst for the memory it runs on.

6 figures
linear probingchainedcuckoo, two tables0240.10.20.30.40.50.60.70.80.9load factorentries read per lookup0120.10.20.30.40.50.60.70.80.9load factorcache misses per lookup8,192 slots, 64 cache lines of 8a probe is an entry read; a miss is a line fetched

Two probes are two misses

Cuckoo hashing's lookup reads at most two slots, and at a load of 0.45 it reads 1.27 on average where linear probing reads 1.39. Replayed through a cache, it misses 1.18 times a lookup where linear probing misses 0.98. The table that wins the count the analysis uses loses the count the machine charges, because two slots in unrelated places are two cache lines, and a run of adjacent slots is usually one.

6 figures
linear probingcuckoo, two tablescuckoo, buckets of 8024680.30.450.60.750.850.95load factorentries read per lookup00.511.50.30.450.60.750.850.95load factorcache misses per lookup8,192 slots, 64 cache lines of 8a probe is an entry read; a miss is a line fetched

The bucket that fits a line

Make each of a cuckoo table's two candidates a bucket of eight slots laid out on one cache line, and no lookup ever touches more than two lines, the table builds past a load of 0.95, and at that load it misses 1.21 times a lookup where linear probing misses 1.79. The prediction that it would lose to linear probing at low loads was wrong — it misses less at every load measured, 0.94 against 0.96 at 0.3 — because a key it holds almost never lives in its second bucket. The guarantee belongs to the alignment, not the bucket; eight slots on lines of four put a lookup on four lines.

6 figures
rounds, if every ready cell ran at oncerow order, stored by rows65,793anti-diagonal order, stored by rows513anti-diagonal order, stored by diagonals513reads that miss the cacherow order, stored by rows6.3%anti-diagonal order, stored by rows31.1%anti-diagonal order, stored by diagonals8.7%fully associative · 32 lines × 8 elements · LRU263,169 reads and writes per order

The table stored the way it is filled

Store an edit-distance table by anti-diagonals instead of by rows, and the anti-diagonal fill keeps its 513 rounds while its cache misses fall from 31.1% of reads to 8.7%. It does not fall to row order's 6.3%, and the gap is not noise — on caches of four and eight lines the two rates are 9.4% and 6.3%, exactly three to two, because a cell reads from two earlier diagonals and only one earlier row. The same layout turns row order into the order that strides, at 28.3%. How a table is stored and the order it is filled in are one decision, and its price is the number of earlier fronts the recurrence reads.

5 figures
124816326410step width, cellscache lines a stepstored by diagonalsby diagonals, each line-alignedstored by rows512 × 512, lines of 8 cellsone step of an anti-diagonal

Eight cells at once

The anti-diagonal fill order exists because its cells do not depend on one another, and every table filled here has been walked one cell at a time anyway. Computed eight at a time, a step touches 5.71 cache lines on the layout that stores the table by diagonals and 10.87 on the one that stores it by rows — and per cell the first keeps falling to 0.42 while the second stops at 1.27. The prediction that a diagonal step would touch three or four lines was wrong, and line-aligning each diagonal only takes it to 4.94.

6 figures

What the libraries do

None of the sorts measured in the other fields is what runs when a program calls sort. A library sort is not an algorithm but a policy — a set of decisions, each with a threshold somebody typed — and the thresholds are where its behaviour actually lives.

Timsort95,770shipsIntrosort130,863shipspdqsort114,408shipsDual-pivot116,836shipsMerge sort96,145textbookHeapsort187,796textbookQuicksort, median-3119,098textbookalgorithmcomparisonsrandom, n = 8,192comparisons, counted exactly

The sort the library ships

Every sorting algorithm measured on this site so far has one thing in common — none of them is what runs when a program calls sort. Python, Java, Rust and Android run Timsort; C++ runs introsort; Java's primitive sort is dual-pivot quicksort. Not one of the four was in this collection, and the reason it matters is that they are not algorithms in the sense the other essays use the word.

7 figures
11010010³10⁴10⁵10⁶10⁷natural runs r in the inputcomparisonsn / minrun = 256TimsortMerge sortInsertionn + n log₂ rn = 8,192, runs built exactlycomparisons, counted exactly

A run is a property of the input

A benchmark that says "nearly sorted" never says how nearly. It is a recipe with a seed, not a measurement, and an adaptive bound stated against it is a bound with an undefined second parameter. Counting the natural runs turns the shape of an input into a number — and then Timsort's bound becomes something that can be fitted rather than quoted.

6 figures
dark: with galloping · pale: with the mode removednearly sorted33,95221,373 jumped−37.7%few distinct values57,91241,178 jumped−36.9%random95,7702 jumped+0.0%n = 8,192, MIN_GALLOP = 7comparisons, counted exactly

When galloping pays

Timsort's merge does not always take elements one at a time. When one run has won seven times in a row it switches to searching for how many to take at once, and switches back when that stops paying. The mode saves 22,104 comparisons on nearly sorted input, 33,270 on input with few distinct values, and costs exactly six on random input — which is the whole design in three numbers.

7 figures
algorithmspread of count ⁄ f(n) — 1.0 is exactTimsortPython, Java objects, Rust, Android1.176nIntrosortC++ std::sort1.104n log nPattern-defeating quicksortRust unstable sort, Boost, libstdc++ 141.155nDual-pivot quicksortJava Arrays.sort, primitives1.624no class grantedtolerance 1.6fitted n = 256–16,384comparisons, counted exactly

The pattern that defeats the pattern

Quicksort's bad cases are patterns — sorted input, organ-pipe input, an adversary's construction. Introsort's answer is to notice the damage and switch algorithms. pdqsort's answer is to notice the pattern and break it, deterministically, with four swaps. On input with eight distinct values that turns a quadratic disaster into a linear sort, and the whole difference is one extra partition scheme.

7 figures
1010010⁵minruncomparisonsshipped: 32randomnearly sortedTimsort, n = 8,192comparisons; rings mark the measured minimum

The threshold somebody chose

A minrun of 32. An insertion cutoff of 16. A gallop threshold of 7. A depth limit of twice the logarithm. Four numbers, in four real source files, none of which appears in any complexity analysis — and each of which decides more about what these algorithms do than the analysis does. Swept, they turn out not to be optima, and finding out what they are instead is the point.

7 figures
H₀ = 3.8916642561k4k16kwindow, in symbolsbits per symbol0.04.38.6LZSSmean symbols covered per match4.65.15.76.98.79.6model: a window of recent text, no probabilities2.13 bits/symbol at a window of 16,384

The dictionary that builds itself

LZSS contains no probability, no frequency table and no entropy calculation. Its entire model is a window of recent text and its only move is to say "the next nine symbols are the ones that appeared 1,200 positions ago". On a stream whose zeroth-order floor is 3.89 bits per symbol it spends 2.11, and widening its window past 4,096 makes it worse rather than better.

7 figures
fades a lazy table computed382,976fades an eager table would compute3,830,256fade multiplications (log scale)11,968 evictions, each comparing all 32 counters at one instant32 counters · half-life 4 s · bursty10× fewer

The fading nobody computes

Twelve thousand arrivals into thirty-two decayed counters cost 382,976 fade multiplications. An implementation that aged every counter on every tick would have cost 3,830,256, and the ratio is exactly the mean gap between arrivals — not a coincidence, and the reason the family is deployable.

5 figures
0%25%50%75%100%1.5×shareheadroom over the mean occupancyoverflowingstanding idledrifting · 4.0 s window · 100 Hz47% overflow at the mean

Sized for a rate that does not hold still

A four-second window on a stream at a hundred arrivals a second holds four hundred items on average and between 105 and 2,169 when the rate moves. An allocation set at that average overflows at 47 per cent of instants while 35 per cent of it stands empty, which is the same decision failing in both directions at once.

9 figures
1,00010,00010³10⁴characters of textcomparisons, built and scannedcrossing at n = 32,000both rules, exactlythe 1979 tables2 patterns of 10 · four symbolscrossing n = 32,000

The rule that pays on a long enough text

With two patterns, the cheap tables cost 106 steps and the scan reads 13,084 characters; the exact tables cost 594 and the scan reads 12,306. Below thirty-two thousand characters the cheap tables win the total, above it the extra skipping pays for them, and with thirty-two patterns there is no crossing at all.

9 figures
1,00010,00010³10⁴characters of textsteps, precomputation plus scancrossing at 8,000published rulesexact rules2 patterns · four symbolscrossing 8,000 · was 32,000

Where the exact rules pay now

With a construction as cheap as the published one, the exact shift rules pay for themselves past eight thousand characters of text at two patterns, four thousand at four, and never at thirty-two — because by thirty-two patterns the two rules make identical decisions.

7 figures
depth 0190.1%depth 19642.9%depth 24,92915.0%depth 37,00921.4%depth 48,47325.9%depth 56,24119.0%depth 63,31010.1%depth 71,2503.8%depth 84221.3%depth 91240.4%depth 10270.1%positions32,768 characters · 1172 phrasesmean 3.98

The number that would choose a cap

A depth histogram is one linear pass — 3.16 operations a character over thirty-two thousand of them — and it says the whole text sits at a mean depth of 3.98 with a worst of ten. Nobody prints it, and every choice of cap in this collection was made without it.

9 figures
100100occurrences of the patternreadsmost frequent ", and " · 388 occurrencesthe scanthe chain12 documents · 144,617 characters3 of 6 won

What the generated collection was right about

Five strands of conclusions, drawn on collections made by one line with one dial, checked against a corpus nobody made. Most hold. One headline was a property of the generator's alphabet, and one crossing that was guessed at turns out to be met — but only with the structure the strand on range minima built.

9 figures
0200400200400occurrences of the patternrows read, and documents in the answerrows readdocuments in the answertwo thousand short documents · 2,214 documents1.04 occurrences a document

The crossing that never arrives

Output-sensitive document listing exists because a pattern can occur four thousand times in eight documents. On a real collection of two thousand short documents it occurs 1.04 times per document, and the whole apparatus buys nothing at all.

7 figures
ten revisions of one filecap 10worst 14 · mean 5.7fourteen revisions of anothercap 10worst 17 · mean 7.2twelve unrelated essayscap 10worst 14 · mean 4.9eight source modulescap 10worst 78 · mean 5.9a generated collectioncap 10worst 13 · mean 4.6within a twentieth of the free parsefour to eight was the model

The cap that would ship

The published sweep put the knee at four to eight. On four real collections it is at nine to twelve, a cap of one costs forty-eight times the free parse rather than twenty-one, and the number a system should actually set is none of those.

8 figures
both halves, one in 3215.0 steps100.0% the sizeforward sampled one in 3215.0 steps83.3% the sizeforward sampled one in 167.0 steps88.4% the sizeforward sampled one in 83.0 steps98.5% the sizeforward sampled one in 41.7 steps118.8% the size8,192 characters · 6 occurrences5.0x faster, 98.5% the size

The saving, spent

A bidirectional index whose reverse half cannot locate is a sixth smaller. Give that sixth back to the half that does locate, and the same total size answers a locate five times faster.

9 figures
a segment tree over the chain2.70x8,142,274 bitsa succinct range minimum over the chain1.62x4,898,356 bitsno chain at all0.84x2,534,726 bitsthe dashed rule is the index itself: a suffix array and the text131,327 characters · 256 documents2.70x → 0.84x

The apparatus that is smaller than its index

Answering "which documents hold this" at a price proportional to the answer used to cost 2.70 times the index it sits beside. Two changes later it costs 0.84, and the largest thing left is an array that says which document each row belongs to.

9 figures
one text16 short documentsprosecopies of one baseprose, one textr/n 0.2996z/n 0.1273prose, many short documentsr/n 0.3053z/n 0.1301copies of one base, run togetherr/n 0.0638z/n 0.0354one copy per documentr/n 0.0768z/n 0.04188,192 characters throughoutthe outlined cell is the one that was empty

A collection is a construction

The same characters, arranged as one copy per document or cut across the copies, give a different run count, a different boundary cost and a different answer about which structure to build. Which one a benchmark used is usually not recorded.

8 figures
bits heldbalanced49,593huffman41,589the answer comes backbalanced: sorted · 26 operationshuffman: sorted · 24 operations32 documents · zipf lengths83.9% of the ordered tree

The smaller tree hands it back unsorted

A frequency-shaped tree is sixteen per cent smaller and returns its documents in code order. A balanced one is larger and returns them sorted. The trade is d log d comparisons against a saving, which is not close — until the answer is truncated.

8 figures
02040608000.50011.502errors permittedbranches that find nothing, per cent5.75x6.98x9.22xthe label is thesaving at that budget8 patterns · 8 characters31.3% to 74.9%

A looser budget wastes a larger share

More errors permitted means more work, and the fraction of that work which was never going to help rises with it — from thirty-one per cent at no errors to seventy-five at two. The saving is worth most where the search is most expensive.

8 figures
0102030100200one sampled position in every …of both halves saved, per centplain marks: 13.5%Elias-Fano: 1.8%16,384 characters17.6x against 2.3x

The floor was the marks

A saving reported as about a sixth of a bidirectional index, falling to an eighth and levelling off. Represent one array properly and it falls to a fiftieth instead — most of what was being dropped was a badly encoded bit vector.

8 figures
a loop over the alphabet107.9the compound walk10.0the cheapest hereone descent per step19.016 exact patterns of 12ranks per step · sigma 21

The saving that is a loss

An operation that is seventy-eight times cheaper on a branching search costs twice as much on an exact one. It reports every symbol present in order to hand back the one that was asked for, and a search that knows its character needs none of the rest.

8 figures
phrases examined per probethe linear scan207.0sorted, with a running maximum10.9 — 19xthe running maximum costs 2,821 bits — 21.6% of the index40 of 40 positions checked, and the two agree at all of them238 phrases · 6,144 characters19x on the scan

The scan the order does not touch

Two hundred and seven phrases examined per probe, against eleven. A running maximum of the source regions' right ends lets a leftward walk stop for good, and it costs twenty-two per cent more bits.

8 figures
10010³10⁴10³10⁴10⁵characters of textoperations, construction includedcrosses at 256the NFAthe DFA64 DFA statescrosses at 256 characters

Where the table starts paying

Five thousand and sixty-five operations before the first character, then one per character. Against nothing before the first character and thirty-nine per character. They cross at two hundred and fifty-six characters, and that crossing is what an engine's compile decision actually is.

8 figures

When the algorithm flips a coin

A skip list's shape is a sequence of coin flips rather than a property of its keys. Where the randomness is the structure rather than a rule applied to one, the distribution is the result and the average is the least of it.