Andrew Noyes
769cf8de9a
This will be used if there's only one check to perform or if the compiler does not support musttail and preserve_none
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
Microbenchmark
Skip list
| ns/op | op/s | err% | ins/op | cyc/op | IPC | bra/op | miss% | total | benchmark |
|---|---|---|---|---|---|---|---|---|---|
| 172.03 | 5,812,791.77 | 0.4% | 3,130.62 | 879.00 | 3.562 | 509.23 | 0.0% | 0.01 | point reads |
| 167.44 | 5,972,130.71 | 0.2% | 3,065.14 | 862.27 | 3.555 | 494.30 | 0.0% | 0.01 | prefix reads |
| 238.77 | 4,188,130.84 | 0.9% | 3,589.93 | 1,259.30 | 2.851 | 637.12 | 0.0% | 0.01 | range reads |
| 424.01 | 2,358,426.70 | 0.2% | 5,620.05 | 2,242.35 | 2.506 | 854.80 | 1.7% | 0.01 | point writes |
| 418.45 | 2,389,780.56 | 0.4% | 5,525.07 | 2,211.05 | 2.499 | 831.71 | 1.7% | 0.01 | prefix writes |
| 254.87 | 3,923,568.88 | 2.6% | 3,187.01 | 1,366.50 | 2.332 | 529.11 | 2.7% | 0.02 | range writes |
| 675.96 | 1,479,374.50 | 3.3% | 7,735.41 | 3,468.60 | 2.230 | 1,386.02 | 1.8% | 0.01 | monotonic increasing point writes |
| 137,986.20 | 7,247.10 | 0.6% | 789,752.33 | 699,462.00 | 1.129 | 144,824.14 | 0.0% | 0.01 | worst case for radix tree |
| 21.63 | 46,231,564.03 | 1.0% | 448.00 | 107.14 | 4.181 | 84.00 | 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 |
|---|---|---|---|---|---|---|---|---|---|
| 12.88 | 77,653,350.77 | 0.5% | 185.37 | 64.45 | 2.876 | 41.51 | 0.4% | 0.01 | point reads |
| 14.67 | 68,179,354.49 | 0.1% | 271.44 | 73.40 | 3.698 | 53.70 | 0.3% | 0.01 | prefix reads |
| 34.84 | 28,701,444.36 | 0.3% | 715.74 | 175.27 | 4.084 | 127.30 | 0.2% | 0.01 | range reads |
| 17.12 | 58,422,988.28 | 0.2% | 314.30 | 86.11 | 3.650 | 39.82 | 0.4% | 0.01 | point writes |
| 31.42 | 31,830,804.65 | 0.1% | 591.06 | 158.07 | 3.739 | 82.67 | 0.2% | 0.01 | prefix writes |
| 37.37 | 26,759,432.70 | 2.2% | 681.98 | 188.95 | 3.609 | 96.10 | 0.1% | 0.01 | range writes |
| 76.72 | 13,035,140.63 | 2.3% | 1,421.28 | 387.17 | 3.671 | 257.76 | 0.1% | 0.01 | monotonic increasing point writes |
| 297,452.00 | 3,361.89 | 0.9% | 3,508,083.00 | 1,500,834.67 | 2.337 | 727,525.33 | 0.1% | 0.01 | worst case for radix tree |
| 87.70 | 11,402,490.60 | 1.0% | 1,795.00 | 442.09 | 4.060 | 297.00 | 0.0% | 0.01 | create and destroy |
"Real data" test
Point queries only, best of three runs. 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: 4.47891 seconds, 364.05 MB/s, Add: 4.55599 seconds, 123.058 MB/s, Gc ratio: 37.1145%
radix tree
Check: 0.953012 seconds, 1710.94 MB/s, Add: 1.30025 seconds, 431.188 MB/s, Gc ratio: 43.9816%, Peak idle memory: 2.28375e+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: 0.804094 seconds, 2027.81 MB/s, Add: 0.652952 seconds, 858.645 MB/s, Gc ratio: 35.3885%