英文:
Creation of concurrency objects dramatically slows down execution time
问题
我已经得到这段代码,并被要求找出如何使用并发来加速处理过程。
如果我运行这段代码,我会得到以下输出:
Elapsed time (us) = 26546
然后我用Go语言编写了一个类似的程序:
package main
import (
"fmt"
"math/rand"
"time"
)
const size int64 = 10000000
var (
a = [size]float32{}
b = [size]float32{}
)
func main() {
var (
i int64
sum float32
time1 time.Time
time2 time.Time
)
rand.Seed(time.Now().UnixNano())
for i = 0; i < size; i++ {
a[i] = rand.Float32()
b[i] = rand.Float32()
}
time1 = time.Now() //Original place
sum = 0.0
for i = 0; i < size; i++ {
sum = sum + a[i] + b[i]
}
time2 = time.Now()
fmt.Printf("Elapsed time (us) = %d\n", time2.Sub(time1).Microseconds())
}
我得到了这个输出(非常令人惊讶地比C版本更快):
Elapsed time (us) = 2462
我的任务是尝试使用并发使其更快,我想到可以通过并行运行数组的创建来加速它们,但是计时器只在创建之后启动。所以我不知道如何加速它,因为值需要合并,这将是一个顺序过程。
所以我将启动计时器移到创建时间之前,得到C程序的结果如下:
Elapsed time (us) = 172496
Go程序的结果如下:
Elapsed time (us) = 247603
所以现在Go比C慢,这是预期的。
然后我尝试将Go程序更改为在每个goroutine中创建每个数组:
package main
import (
"fmt"
"math/rand"
"sync"
"time"
)
const size int = 10000000
var (
a = [size]float64{}
b = [size]float64{}
)
func main() {
var (
wg sync.WaitGroup
sum float64
time1 time.Time
time2 time.Time
)
rand.Seed(time.Now().UnixNano())
wg.Add(2)
time1 = time.Now()
go func() {
for i := 0; i < size; i++ {
a[i] = rand.Float64()
}
wg.Done()
}()
go func() {
for i := 0; i < size; i++ {
b[i] = rand.Float64()
}
wg.Done()
}()
wg.Wait()
sum = 0.0
for i := 0; i < size; i++ {
sum = sum + a[i] + b[i]
}
time2 = time.Now()
fmt.Printf("Elapsed time (us) = %d\n", time2.Sub(time1).Microseconds())
}
我得到了以下输出:
Elapsed time (us) = 395808
这相当慢,我认为这与函数的调用和等待组逻辑有关。
然后我尝试使用通道。
这只会使程序运行很长时间,并且代码非常冗长。
然后我尝试让每个goroutine自己添加字段:
package main
import (
"fmt"
"math/rand"
"sync"
"time"
)
const size int = 10000000
func main() {
var (
wg sync.WaitGroup
sum float64
asum float64
bsum float64
time1 time.Time
time2 time.Time
)
rand.Seed(time.Now().UnixNano())
wg.Add(2)
time1 = time.Now()
go func() {
asum = 0
for i := 0; i < size; i++ {
asum = asum + rand.Float64()
}
wg.Done()
}()
go func() {
bsum = 0
for i := 0; i < size; i++ {
bsum = bsum + rand.Float64()
}
wg.Done()
}()
wg.Wait()
sum = asum + bsum
time2 = time.Now()
fmt.Printf("Elapsed time (us) = %d\n", time2.Sub(time1).Microseconds())
fmt.Println(sum)
}
它返回了以下结果:
Elapsed time (us) = 395182
1.000137482475232e+07
我必须使用sum变量才能运行程序-这就是为什么我打印它。
所以我似乎无法通过并发使这个程序运行得更快。
有人对此有什么提示吗?或者我应该在并发产生效果之前运行更多的任务?这只是因为我在这种情况下只处理了2个任务,并且数组处理速度非常快吗?
英文:
I have gotten this code and been asked to find out how I can use concurrency to speed up the process.
#include <stdio.h>
#include <stdlib.h>
#include <time.h>
#include <sys/time.h>
#define SIZE 10000000
volatile float a[SIZE];
volatile float b[SIZE];
int main(int argc, char **argv)
{
long int i;
double sum;
struct timeval time1, time2;
srand(time(0));
for (i = 0; i < SIZE; i++)
{
a[i] = rand();
b[i] = rand();
}
gettimeofday(&time1, 0); //Original place
sum = 0.0;
for (i = 0; i < SIZE; i++)
{
sum = sum + a[i]*b[i];
}
gettimeofday(&time2, 0);
printf("Elapsed time (us) = %d\n", (time2.tv_sec-time1.tv_sec)*1000000 + time2.tv_usec - time1.tv_usec);
return 0;
}
if I run the code I get the output
Elapsed time (us) = 26546
Then I wrote a similar program in Go
package main
import (
"fmt"
"math/rand"
"time"
)
const size int64 = 10000000
var (
a = [size]float32{}
b = [size]float32{}
)
func main() {
var (
i int64
sum float32
time1 time.Time
time2 time.Time
)
rand.Seed(time.Now().UnixNano())
for i = 0; i < size; i++ {
a[i] = rand.Float32()
b[i] = rand.Float32()
}
time1 = time.Now() //Original place
sum = 0.0
for i = 0; i < size; i++ {
sum = sum + a[i] + b[i]
}
time2 = time.Now()
fmt.Printf("Elapsed time (us) = %d\n", time2.Sub(time1).Microseconds())
}
An I get this output (which was very surprisingly faster than the C version)
Elapsed time (us) = 2462
My job was to try to make it faster with concurrency, and I was thinking that the creation of the arrays could be speed up if they would be run in parallel, However the timer is only started after the creation. So then I don't really know how I can speed it up since the values need to be merges which would be a sequential process.
So I move the start timer over the creation time and get for the c program:
Elapsed time (us) = 172496
and for the go program:
Elapsed time (us) = 247603
So now go is slower than C as expected.
Then I tried to change my go program to create each array in its own goroutine:
package main
import (
"fmt"
"math/rand"
"sync"
"time"
)
const size int = 10000000
var (
a = [size]float64{}
b = [size]float64{}
)
func main() {
var (
wg sync.WaitGroup
sum float64
time1 time.Time
time2 time.Time
)
rand.Seed(time.Now().UnixNano())
wg.Add(2)
time1 = time.Now()
go func() {
for i := 0; i < size; i++ {
a[i] = rand.Float64()
}
wg.Done()
}()
go func() {
for i := 0; i < size; i++ {
b[i] = rand.Float64()
}
wg.Done()
}()
wg.Wait()
sum = 0.0
for i := 0; i < size; i++ {
sum = sum + a[i] + b[i]
}
time2 = time.Now()
fmt.Printf("Elapsed time (us) = %d\n", time2.Sub(time1).Microseconds())
}
and I get the output:
Elapsed time (us) = 395808
Which is quite slow. and I expect that this has something to do with the invokation of the functions and the waitgroup logic.
Then I tried with channels.
Which just made the program take forever, and the code waay to long.
Then I tried with each coroutine adding the fields itself
package main
import (
"fmt"
"math/rand"
"sync"
"time"
)
const size int = 10000000
func main() {
var (
wg sync.WaitGroup
sum float64
asum float64
bsum float64
time1 time.Time
time2 time.Time
)
rand.Seed(time.Now().UnixNano())
wg.Add(2)
time1 = time.Now()
go func() {
asum = 0
for i := 0; i < size; i++ {
asum = asum + rand.Float64()
}
wg.Done()
}()
go func() {
bsum = 0
for i := 0; i < size; i++ {
bsum = bsum + rand.Float64()
}
wg.Done()
}()
wg.Wait()
sum = asum + bsum
time2 = time.Now()
fmt.Printf("Elapsed time (us) = %d\n", time2.Sub(time1).Microseconds())
fmt.Println(sum)
}
which returned
Elapsed time (us) = 395182
1.000137482475232e+07
I had to use the sum as well to be able to run the program - thats why I print it.
So I just cant seem to get this program to run any faster with concurrency.
Does anyone have a hint for me? or should I just run more jobs before concurrency will have any effect? Is it just because I only deal with 2 jobs in this case, and because arrays are so fast to process?
答案1
得分: 2
并发可以加快执行时间。
Go程序:
经过的时间(微秒)= 130768
带有并发的Go程序:
经过的时间(微秒)= 66947
为了使每个goroutine都拥有自己的rand.Rand实例,请使用rand.New(src Source)。
运行C程序的Go版本。
x.go
:
package main
import (
"fmt"
"math/rand"
"time"
)
const size = 10000000
var (
a = [size]float32{}
b = [size]float32{}
)
func main() {
start := time.Now()
r := rand.New(rand.NewSource(time.Now().UnixNano()))
for i := 0; i < size; i++ {
a[i] = r.Float32()
b[i] = r.Float32()
}
sum := 0.0
for i := 0; i < size; i++ {
sum += float64(a[i]) * float64(b[i])
}
since := time.Since(start).Microseconds()
fmt.Printf("经过的时间(微秒)= %d\n", since)
}
.
$ go build x.go && ./x
经过的时间(微秒)= 130768
$
运行并发的Go版本的C程序。
y.go
:
package main
import (
"fmt"
"math/rand"
"sync"
"time"
)
const size = 10000000
var (
a = [size]float32{}
b = [size]float32{}
)
func main() {
start := time.Now()
var wg sync.WaitGroup
wg.Add(2)
go func() {
defer wg.Done()
r := rand.New(rand.NewSource(time.Now().UnixNano()))
for i := 0; i < size; i++ {
a[i] = r.Float32()
}
}()
go func() {
defer wg.Done()
r := rand.New(rand.NewSource(time.Now().UnixNano()))
for i := 0; i < size; i++ {
b[i] = r.Float32()
}
}()
wg.Wait()
sum := 0.0
for i := 0; i < size; i++ {
sum += float64(a[i]) * float64(b[i])
}
since := time.Since(start).Microseconds()
fmt.Printf("经过的时间(微秒)= %d\n", since)
}
.
$ go build y.go && ./y
经过的时间(微秒)= 66947
$
英文:
Concurrency speeds up execution time.
Go program:
Elapsed time (us) = 130768
Go program with concurrency:
Elapsed time (us) = 66947
For each goroutine to have its own rand.Rand instance, use rand.New(src Source).
Run a Go version of the C program.
x.go
:
package main
import (
"fmt"
"math/rand"
"time"
)
const size = 10000000
var (
a = [size]float32{}
b = [size]float32{}
)
func main() {
start := time.Now()
r := rand.New(rand.NewSource(time.Now().UnixNano()))
for i := 0; i < size; i++ {
a[i] = r.Float32()
b[i] = r.Float32()
}
sum := 0.0
for i := 0; i < size; i++ {
sum += float64(a[i]) * float64(b[i])
}
since := time.Since(start).Microseconds()
fmt.Printf("Elapsed time (us) = %d\n", since)
}
.
$ go build x.go && ./x
Elapsed time (us) = 130768
$
Run a concurrent Go version of the C program.
y.go
:
package main
import (
"fmt"
"math/rand"
"sync"
"time"
)
const size = 10000000
var (
a = [size]float32{}
b = [size]float32{}
)
func main() {
start := time.Now()
var wg sync.WaitGroup
wg.Add(2)
go func() {
defer wg.Done()
r := rand.New(rand.NewSource(time.Now().UnixNano()))
for i := 0; i < size; i++ {
a[i] = r.Float32()
}
}()
go func() {
defer wg.Done()
r := rand.New(rand.NewSource(time.Now().UnixNano()))
for i := 0; i < size; i++ {
b[i] = r.Float32()
}
}()
wg.Wait()
sum := 0.0
for i := 0; i < size; i++ {
sum += float64(a[i]) * float64(b[i])
}
since := time.Since(start).Microseconds()
fmt.Printf("Elapsed time (us) = %d\n", since)
}
.
$ go build y.go && ./y
Elapsed time (us) = 66947
$
通过集体智慧和协作来改善编程学习和解决问题的方式。致力于成为全球开发者共同参与的知识库,让每个人都能够通过互相帮助和分享经验来进步。
评论