On Thursday, January 26, 2017 at 11:15:39 PM UTC+8, andrey mirtchovski
wrote:
>
> I bet you get wildly differing results with each run. That's an
> indicator that your sample size is not statistically relevant (there's
> randomness introduced both in the select, as well as in the OS and Go
> schedulers).
>
> Increase the iterations until your results stabilize, then you'll see
> what the real behaviour is:
>
> $ for i in 100000 1000000 10000000 100000000 1000000000; do echo $i;
> go run t.go $i; done
> 100000
> pass one: 48154/51846=0.93
> pass two: 41040/58960=0.70
> pass three: 54133/45867=1.18
> 1000000
> pass one: 531420/468580=1.13
> pass two: 671204/328796=2.04
> pass three: 481769/518231=0.93
> 10000000
> pass one: 4983060/5016940=0.99
> pass two: 5475692/4524308=1.21
> pass three: 4865659/5134341=0.95
> 100000000
> pass one: 50361915/49638085=1.01
> pass two: 54589075/45410925=1.20
> pass three: 47290086/52709914=0.90
>
The result for 1000000000 is the same as 300000 on my machine.
It looks the reason is starvation (see my last comment).
>
> On Thu, Jan 26, 2017 at 5:06 AM, T L <tapi...@gmail.com <javascript:>>
> wrote:
> >
> > package main
> >
> > import "fmt"
> >
> > func main() {
> > c0, c1 := make(chan int, 10), make(chan int, 10)
> >
> > go func() {
> > for {
> > c0 <- 0
> > }
> > }()
> >
> > go func() {
> > for {
> > c1 <- 1
> > }
> > }()
> >
> > var n0, n1 int
> >
> > f0 := func() {
> > n0++
> > }
> >
> > f1 := func() {
> > n1++
> > }
> >
> > // pass one
> >
> > n0, n1 = 0, 0
> > for range [300000]struct{}{} {
> > select {
> > case <- c0: f0()
> > case <- c1: f1()
> > }
> > }
> >
> > fmt.Printf("pass one: %d/%d=%.2f \n", n0, n1, float64(n0) /
> float64(n1))
> >
> > // pass two
> >
> > n0, n1 = 0, 0
> > for range [300000]struct{}{} {
> > select {
> > case <- c0: f0()
> > case <- c1: f1()
> > case <- c1: f1()
> > }
> > }
> >
> > fmt.Printf("pass two: %d/%d=%.2f \n", n0, n1, float64(n0) /
> float64(n1))
> >
> > // pass three
> >
> > n0, n1 = 0, 0
> > for range [300000]struct{}{} {
> > select {
> > case <- c0: f0()
> > case <- c0: f0()
> > case <- c1: f1()
> > }
> > }
> >
> > fmt.Printf("pass three: %d/%d=%.2f \n", n0, n1, float64(n0) /
> > float64(n1))
> > }
> >
> > The output:
> >
> > pass one: 154970/145030=1.07
> > pass two: 177853/122147=1.46
> > pass three: 127768/172232=0.74
> >
> > --
> > You received this message because you are subscribed to the Google
> Groups
> > "golang-nuts" group.
> > To unsubscribe from this group and stop receiving emails from it, send
> an
> > email to golang-nuts...@googlegroups.com <javascript:>.
> > For more options, visit https://groups.google.com/d/optout.
>
--
You received this message because you are subscribed to the Google Groups
"golang-nuts" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to golang-nuts+unsubscr...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.
