See https://kumagi.hatenablog.com/entry/ring-buffer
| Type | Desc. |
|---|---|
| RingBuffer0 | Naive implementation with write and read indexes with modulo |
| RingBuffer1 | Use bit and op instead of modulo (based RingBuffer0) |
| RingBuffer2 | Support goroutine with sync.Mutex (based RingBuffer1) |
| RingBuffer3 | Use sync/atomic for multi-goroutine (based RingBuffer1) |
| RingBuffer3B | Same as RingBuffer3 but w/o index caches |
| RingBuffer4 | Use chan for multi-goroutine |
| RingBuffer4B | Use chan for multi-goroutine with blocking |
| RingBuffer5 | Use read index and capacity length w/o multi-goroutine supports |
goos: windows
goarch: amd64
pkg: github.com/koron/go-ringbuf-benchmark
cpu: Intel(R) Core(TM) i9-9900K CPU @ 3.60GHz
BenchmarkRingBuffer0/single-16 178371265 6.526 ns/op
BenchmarkRingBuffer1/single-16 786034266 1.505 ns/op
BenchmarkRingBuffer2/single-16 85309674 12.78 ns/op
BenchmarkRingBuffer2/multi-16 69087810 16.26 ns/op
BenchmarkRingBuffer3/single-16 140011688 8.641 ns/op
BenchmarkRingBuffer3/multi-16 37608711 30.85 ns/op
BenchmarkRingBuffer3B/single-16 139669322 8.494 ns/op
BenchmarkRingBuffer3B/multi-16 56262482 33.03 ns/op
BenchmarkRingBuffer4/single-16 57601750 20.16 ns/op
BenchmarkRingBuffer4/multi-16 57848888 21.95 ns/op
BenchmarkRingBuffer4B/single-16 58735713 19.57 ns/op
BenchmarkRingBuffer4B/multi-16 55689105 21.48 ns/op
BenchmarkRingBuffer5/single-16 665201374 1.753 ns/op
PASS
ok github.com/koron/go-ringbuf-benchmark 19.022s- 複数goroutine環境での評価
- sync.Mutex(2)は意外と安定
- sync/atomic(3)は性能劣化が著しい
- chan(4)で良いのでは?
- 1 goroutineでの評価
- index+書き込まれ量の管理方式(5)は、意外と良い。ただし複数goroutine化の見通しが立たないか、もしくは性能が悪そう。