A data structure for optimistic concurrency control on ranges of bitwise-lexicographically-ordered keys.
Intended as an alternative to FoundationDB's skip list.
Hardware for all benchmarks is an AMD Ryzen 9 7900 with (2x32GB) 5600MT/s CL28-34-34-89 1.35V RAM.
$ clang++ --version
Ubuntu clang version 20.0.0 (++20241119082716+7e85cb8a8a9d-1~exp1~20241119082825.551)
Target: x86_64-pc-linux-gnu
Thread model: posix
InstalledDir: /usr/lib/llvm-20/bin
Microbenchmark
Skip list
| ns/op | op/s | err% | ins/op | cyc/op | IPC | bra/op | miss% | total | benchmark |
|---|---|---|---|---|---|---|---|---|---|
| 175.88 | 5,685,575.81 | 0.0% | 3,014.03 | 883.13 | 3.413 | 504.59 | 0.0% | 2.10 | point reads |
| 173.35 | 5,768,718.52 | 0.0% | 2,954.16 | 869.59 | 3.397 | 490.17 | 0.0% | 2.07 | prefix reads |
| 251.74 | 3,972,274.68 | 0.1% | 3,592.41 | 1,265.71 | 2.838 | 629.31 | 0.0% | 3.01 | range reads |
| 472.13 | 2,118,079.83 | 0.2% | 4,450.57 | 2,377.84 | 1.872 | 707.92 | 2.3% | 5.60 | point writes |
| 471.34 | 2,121,609.91 | 0.0% | 4,410.22 | 2,372.55 | 1.859 | 694.74 | 2.3% | 5.61 | prefix writes |
| 306.69 | 3,260,650.68 | 0.0% | 2,315.38 | 1,550.33 | 1.493 | 396.69 | 3.3% | 3.68 | range writes |
| 502.70 | 1,989,277.15 | 0.9% | 6,999.33 | 2,527.47 | 2.769 | 1,251.74 | 1.3% | 0.06 | monotonic increasing point writes |
| 138,097.67 | 7,241.25 | 0.7% | 807,445.67 | 699,899.00 | 1.154 | 144,584.50 | 0.8% | 0.01 | worst case for radix tree |
| 46.29 | 21,605,126.00 | 1.0% | 902.00 | 230.73 | 3.909 | 132.00 | 0.0% | 0.01 | create and destroy |
Radix tree (this implementation)
| ns/op | op/s | err% | ins/op | cyc/op | IPC | bra/op | miss% | total | benchmark |
|---|---|---|---|---|---|---|---|---|---|
| 14.04 | 71,216,855.01 | 0.1% | 245.98 | 70.38 | 3.495 | 31.49 | 0.6% | 0.17 | point reads |
| 14.57 | 68,630,592.89 | 0.1% | 297.83 | 72.93 | 4.084 | 40.34 | 0.5% | 0.17 | prefix reads |
| 35.28 | 28,347,588.39 | 0.2% | 783.70 | 176.57 | 4.438 | 107.65 | 0.2% | 0.42 | range reads |
| 20.64 | 48,453,280.08 | 0.0% | 378.97 | 103.31 | 3.668 | 49.92 | 0.7% | 0.25 | point writes |
| 39.14 | 25,551,096.17 | 0.0% | 672.00 | 196.08 | 3.427 | 101.28 | 0.3% | 0.47 | prefix writes |
| 39.88 | 25,075,121.85 | 0.0% | 738.26 | 199.71 | 3.697 | 111.59 | 0.1% | 0.49 | range writes |
| 79.36 | 12,601,312.65 | 0.5% | 1,447.65 | 398.77 | 3.630 | 275.61 | 0.1% | 0.01 | monotonic increasing point writes |
| 307,037.33 | 3,256.93 | 0.3% | 4,043,060.50 | 1,549,199.33 | 2.610 | 714,828.00 | 0.1% | 0.01 | worst case for radix tree |
| 97.00 | 10,309,246.47 | 0.2% | 1,986.00 | 487.06 | 4.078 | 315.00 | 0.0% | 0.01 | create and destroy |
"Real data" test
Point queries only. Gc ratio is the ratio of time spent doing garbage collection to time spent adding writes or doing garbage collection. Lower is better.
skip list
Check: 4.66163 seconds, 361.716 MB/s, Add: 3.97534 seconds, 144.726 MB/s, Gc ratio: 33.3998%, Peak idle memory: 5.61007e+06
radix tree
Check: 1.01776 seconds, 1656.77 MB/s, Add: 1.2718 seconds, 452.38 MB/s, Gc ratio: 38.626%, Peak idle memory: 2.05667e+06
hash table
(The hash table implementation doesn't work on range queries, and its purpose is to provide an idea of how fast point queries can be)
Check: 0.859188 seconds, 1962.54 MB/s, Add: 0.714174 seconds, 805.596 MB/s, Gc ratio: 34.9552%, Peak idle memory: 0