[go語言]Buffered channel、slice和mutex的簡單效能測試

這是一個建立於 的文章，其中的資訊可能已經有所發展或是發生改變。

測試代碼：

package mainimport (    "fmt"    "runtime"    "sync"    "time")const COUNT = 1000000func bench1(ch chan int) time.Duration {    t := time.Now()    for i := 0; i < COUNT; i++ {        ch <- i    }    var v int    for i := 0; i < COUNT; i++ {        v = <-ch    }    _ = v    return time.Now().Sub(t)}func bench2(s []int) time.Duration {    t := time.Now()    for i := 0; i < COUNT; i++ {        s[i] = i    }    var v int    for i := 0; i < COUNT; i++ {        v = s[i]    }    _ = v    return time.Now().Sub(t)}func bench3(s []int, mutex *sync.Mutex) time.Duration {    t := time.Now()    for i := 0; i < COUNT; i++ {        mutex.Lock()        s[i] = i        mutex.Unlock()    }    var v int    for i := 0; i < COUNT; i++ {        mutex.Lock()        v = s[i]        mutex.Unlock()    }    _ = v    return time.Now().Sub(t)}func main() {    runtime.GOMAXPROCS(runtime.NumCPU())    ch := make(chan int, COUNT)    s := make([]int, COUNT)    var mutex sync.Mutex    fmt.Println("channel\tslice\tmutex_slice")    for i := 0; i < 10; i++ {        fmt.Printf("%v\t%v\t%v\n", bench1(ch), bench2(s), bench3(s, &mutex))    }}

測試環境CPU: i7-3770MEMORY: 32GOS: ubuntu12.04 x86_64GO VERSION: 1.0.3
輸出：

channel	slice	mutex_slice
53.774ms	1.735ms	37.103ms
52.978ms	1.058ms	36.928ms
52.864ms	1.058ms	36.928ms
53.337ms	1.069ms	38.073ms
53.695ms	1.055ms	37.801ms
53.45ms	1.063ms	37.683ms
53.678ms	1.161ms	37.767ms
3.568ms	1.052ms	37.792ms
53.47ms	1.06ms	37.185ms
52.78ms	1.062ms	36.899ms

結論：在沒有競爭的情況下，緩衝通道比線程鎖稍慢，但執行時間是直接對數組讀寫的40~50倍。引申：對於之前的提到的記憶體 Clerk，或者其他類型的資源分派器，如果頻繁調用的話，還是限制在goroutine記憶體 Clerk更合適。應該盡量避免在goroutine間分配資源。當然，實際的效能調整應該基於profile定位效能瓶頸而不是單純的想象。