Andrew Noyes
2e08b54785
Some checks reported errors
weaselab/conflict-set/pipeline/head Something is wrong with the build of this commit
Closes #7
106 lines
3.7 KiB
Markdown
106 lines
3.7 KiB
Markdown
A data structure for optimistic concurrency control on ranges of bitwise-lexicographically-ordered keys.
|
|
|
|
Intended to replace FoundationDB's skip list.
|
|
|
|
Hardware for all benchmarks is a mac m1 2020.
|
|
|
|
# FoundationDB's benchmark
|
|
|
|
## Skip list
|
|
|
|
```
|
|
New conflict set: 1.964 sec
|
|
0.637 Mtransactions/sec
|
|
2.546 Mkeys/sec
|
|
Detect only: 1.859 sec
|
|
0.672 Mtransactions/sec
|
|
2.690 Mkeys/sec
|
|
Skiplist only: 1.275 sec
|
|
0.980 Mtransactions/sec
|
|
3.921 Mkeys/sec
|
|
Performance counters:
|
|
Build: 0.0496
|
|
Add: 0.0539
|
|
Detect: 1.86
|
|
D.Sort: 0.412
|
|
D.Combine: 0.0139
|
|
D.CheckRead: 0.682
|
|
D.CheckIntraBatch: 0.00673
|
|
D.MergeWrite: 0.593
|
|
D.RemoveBefore: 0.148
|
|
```
|
|
|
|
## Radix tree (this implementation)
|
|
|
|
```
|
|
New conflict set: 1.289 sec
|
|
0.970 Mtransactions/sec
|
|
3.879 Mkeys/sec
|
|
Detect only: 1.199 sec
|
|
1.043 Mtransactions/sec
|
|
4.170 Mkeys/sec
|
|
Skiplist only: 0.542 sec
|
|
2.305 Mtransactions/sec
|
|
9.220 Mkeys/sec
|
|
Performance counters:
|
|
Build: 0.0395
|
|
Add: 0.0492
|
|
Detect: 1.2
|
|
D.Sort: 0.411
|
|
D.Combine: 0.0135
|
|
D.CheckRead: 0.22
|
|
D.CheckIntraBatch: 0.00652
|
|
D.MergeWrite: 0.322
|
|
D.RemoveBefore: 0.223
|
|
```
|
|
|
|
# Our benchmark
|
|
|
|
## Skip list
|
|
|
|
| ns/op | op/s | err% | total | benchmark
|
|
|--------------------:|--------------------:|--------:|----------:|:----------
|
|
| 249.47 | 4,008,533.72 | 0.5% | 0.01 | `point reads`
|
|
| 236.78 | 4,223,252.12 | 0.4% | 0.01 | `prefix reads`
|
|
| 432.76 | 2,310,737.74 | 0.1% | 0.01 | `range reads`
|
|
| 449.42 | 2,225,105.96 | 0.5% | 0.01 | `point writes`
|
|
| 438.47 | 2,280,674.39 | 0.3% | 0.01 | `prefix writes`
|
|
| 242.92 | 4,116,581.59 | 3.7% | 0.02 | `range writes`
|
|
| 553.31 | 1,807,319.60 | 0.7% | 0.01 | `monotonic increasing point writes`
|
|
|
|
## Radix tree (this implementation)
|
|
|
|
| ns/op | op/s | err% | total | benchmark
|
|
|--------------------:|--------------------:|--------:|----------:|:----------
|
|
| 19.35 | 51,669,510.11 | 0.2% | 0.01 | `point reads`
|
|
| 56.64 | 17,655,057.00 | 0.2% | 0.01 | `prefix reads`
|
|
| 217.04 | 4,607,450.05 | 0.2% | 0.01 | `range reads`
|
|
| 26.31 | 38,012,015.29 | 0.0% | 0.01 | `point writes`
|
|
| 41.45 | 24,124,003.19 | 0.2% | 0.01 | `prefix writes`
|
|
| 48.33 | 20,691,082.14 | 0.0% | 0.01 | `range writes`
|
|
| 84.92 | 11,775,856.81 | 4.0% | 0.01 | `monotonic increasing point writes`
|
|
|
|
# "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: 11.267 seconds, 331.812 MB/s, Add: 5.33323 seconds, 131.635 MB/s, Gc ratio: 45.4671%
|
|
```
|
|
|
|
## radix tree
|
|
|
|
```
|
|
Check: 2.48508 seconds, 1504.38 MB/s, Add: 2.07295 seconds, 338.666 MB/s, Gc ratio: 42.2825%
|
|
```
|
|
|
|
## 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: 1.83931 seconds, 2032.56 MB/s, Add: 0.607861 seconds, 1154.93 MB/s, Gc ratio: 48.7142%
|
|
```
|