Goroutine & bufferred channel's quick performance taste: TPS from 28k to 16m

Objective

To get TPS (Transaction Per Second) as high as possible by using goroutine & bufferred channel (CSP pattern)

Environment

2.3 GHz Intel Core i5; 8 GB 2133 MHz LPDDR3
macOS Mojave 10.14.4
Go 1.12.6

Setup Go

If you are new to Golang, for Go’s setup, please refer to my blog: Go - Build a web API

Create a Go project

mkdir -p ~/ws/golang/src/blue/demo
cd ~/ws/golang/src/blue/demo
vim demo.go

package main

import (
  "fmt"
  "time"
)

var load = 100000

// CSP Communicating Sequential Processes
func consumer(data chan int, done chan bool) {
  for i := range data {
    fmt.Println("Consuming:", i)
  }

  done <- true
}

func producer(data chan int) {
  for i := 0; i < load; i++ {
    fmt.Println("Producing:", i)
    data <- i
  }
  close(data)
}

func demoChannelRoutine() {
  start := time.Now()
  data := make(chan int)
  done := make(chan bool, 1)

  go consumer(data, done)
  go producer(data)

  <-done

  elapsed := time.Since(start).Seconds()
  fmt.Printf("Elapsed: %.3fs\n", elapsed)
  tps := float64(load) / elapsed
  fmt.Printf("TPS: %.0f\n", tps)
}

1st run, TPS 28k

go build demo.go
./demo

# Output
...
Producing: 99998
Producing: 99999
Consuming: 99998
Consuming: 99999
Elapsed: 3.506s
Transaction Per Second: 28521
Cleaning up..

Removing printing I/O and 2nd run, TPS boosts from 28k to 3m


// Only changed part listed here
var load = 100000

func consumer(data chan int, done chan bool) {
  for i := range data {
    //fmt.Println("Consuming:", i)

    if i < 0 {
    }
  }

  done <- true
}

func producer(data chan int) {
  for i := 0; i < load; i++ {
    //fmt.Println("Producing:", i)
    data <- i
  }
  close(data)
}

go build demo.go
./demo

# Output
...
Elapsed: 0.027s
Transaction Per Second: 3738271
Cleaning up..

Bufferred channel & 3rd run, TPS boosts from 3m to 16m

// Only changed part listed here

var load = 100000000
var chBuf = 10000
...

data := make(chan int, chBuf)

 go build demo.go
./demo

# Output
Elapsed: 6.047s
Transaction Per Second: 16537191
Cleaning up..

Conclusion

This is a quick test of goroutine + bufferred channel. From introducing heavy I/O (printing) to only RAM I/O. I could boost TPS from 28k to 3m. By introducing bufferred channel, I boosted TPS from 3m to 16m.