4.6 KiB
4.6 KiB
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 (++20241118082208+63b926af5ff4-1~exp1~20241118082226.549)
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
|--------------------:|--------------------:|--------:|----------------:|----------------:|-------:|---------------:|--------:|----------:|:----------
| 164.92 | 6,063,693.08 | 0.0% | 3,014.03 | 829.48 | 3.634 | 504.59 | 0.0% | 1.97 | `point reads`
| 162.85 | 6,140,505.89 | 0.0% | 2,954.16 | 819.09 | 3.607 | 490.17 | 0.0% | 1.94 | `prefix reads`
| 243.06 | 4,114,132.65 | 0.0% | 3,592.41 | 1,224.74 | 2.933 | 629.31 | 0.0% | 2.90 | `range reads`
| 455.46 | 2,195,561.38 | 0.0% | 4,450.57 | 2,301.93 | 1.933 | 707.92 | 2.1% | 5.44 | `point writes`
| 454.16 | 2,201,858.08 | 0.0% | 4,410.22 | 2,295.25 | 1.921 | 694.74 | 2.1% | 5.42 | `prefix writes`
| 302.97 | 3,300,699.07 | 0.0% | 2,315.38 | 1,531.20 | 1.512 | 396.69 | 3.3% | 3.64 | `range writes`
| 493.69 | 2,025,564.75 | 0.9% | 6,999.33 | 2,493.53 | 2.807 | 1,251.74 | 1.3% | 0.06 | `monotonic increasing point writes`
| 136,298.50 | 7,336.84 | 0.9% | 807,444.50 | 693,845.50 | 1.164 | 144,584.50 | 0.9% | 0.01 | `worst case for radix tree`
| 47.68 | 20,974,738.11 | 0.7% | 902.00 | 238.47 | 3.783 | 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
|--------------------:|--------------------:|--------:|----------------:|----------------:|-------:|---------------:|--------:|----------:|:----------
| 12.73 | 78,578,520.69 | 0.0% | 247.13 | 64.01 | 3.861 | 32.64 | 0.6% | 0.15 | `point reads`
| 14.48 | 69,077,363.01 | 0.1% | 299.99 | 72.83 | 4.119 | 42.50 | 0.4% | 0.17 | `prefix reads`
| 35.33 | 28,307,787.32 | 0.1% | 782.70 | 177.67 | 4.405 | 106.65 | 0.2% | 0.42 | `range reads`
| 20.64 | 48,445,877.80 | 0.0% | 376.04 | 103.84 | 3.621 | 49.97 | 0.7% | 0.25 | `point writes`
| 38.32 | 26,098,688.72 | 0.0% | 665.25 | 192.72 | 3.452 | 101.33 | 0.4% | 0.46 | `prefix writes`
| 39.46 | 25,343,458.68 | 0.0% | 732.48 | 198.44 | 3.691 | 111.75 | 0.1% | 0.48 | `range writes`
| 80.63 | 12,401,923.78 | 2.4% | 1,461.96 | 407.42 | 3.588 | 278.93 | 0.1% | 0.01 | `monotonic increasing point writes`
| 311,077.67 | 3,214.63 | 0.5% | 4,016,995.00 | 1,582,232.00 | 2.539 | 714,572.00 | 0.1% | 0.01 | `worst case for radix tree`
| 106.36 | 9,401,962.15 | 0.9% | 2,046.00 | 538.76 | 3.798 | 329.00 | 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: 4.35798 seconds, 386.919 MB/s, Add: 3.69297 seconds, 155.792 MB/s, Gc ratio: 33.6285%, Peak idle memory: 5.61007e+06
```
## radix tree
```
Check: 1.0184 seconds, 1655.72 MB/s, Add: 1.3866 seconds, 414.924 MB/s, Gc ratio: 34.8706%, Peak idle memory: 2.32922e+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: 0.855065 seconds, 1972 MB/s, Add: 0.711856 seconds, 808.219 MB/s, Gc ratio: 34.9436%, Peak idle memory: 0
```