Benchmarking sync.RWMutex vs. sync/atomic Load/Store functions in Go 1.4

mutexvsatomic.go

package mutexvsatomic_test

import (
	"sync"
	"sync/atomic"
	"testing"
)

var a int32

const (
	READS  = 000
	WRITES = 100
)

var readmu = func(mu *sync.RWMutex) {
	for x := 0; x < READS; x++ {
		mu.RLock()
		_ = a
		mu.RUnlock()
	}
}
var writemu = func(mu *sync.RWMutex) {
	for x := 0; x < WRITES; x++ {
		mu.Lock()
		a = int32(x)
		mu.Unlock()
	}
}

var read = func() {
	for x := 0; x < READS; x++ {
		_ = atomic.LoadInt32(&a)
	}
}

var write = func() {
	for x := 0; x < WRITES; x++ {
		atomic.StoreInt32(&a, int32(x))
	}
}

var readuns = func() {
	for x := 0; x < READS; x++ {
		_ = a
	}
}
var writeuns = func() {
	for x := 0; x < WRITES; x++ {
		a = int32(x)
	}
}

func BenchmarkMutex(b *testing.B) {
	b.RunParallel(func(pb *testing.PB) {
		for pb.Next() {
			wg := sync.WaitGroup{}
			mu := sync.RWMutex{}
			go func() {
				wg.Add(1)
				writemu(&mu)
				wg.Done()
			}()
			go func() {
				wg.Add(1)
				readmu(&mu)
				wg.Done()
			}()
			wg.Wait()
		}
	})

}

func BenchmarkAtomic(b *testing.B) {
	b.RunParallel(func(pb *testing.PB) {
		for pb.Next() {
			wg := sync.WaitGroup{}
			go func() {
				wg.Add(1)
				write()
				wg.Done()
			}()
			go func() {
				wg.Add(1)
				read()
				wg.Done()
			}()
			wg.Wait()
		}
	})

}

func BenchmarkUnsafe(b *testing.B) {
	b.RunParallel(func(pb *testing.PB) {
		for pb.Next() {
			wg := sync.WaitGroup{}
			go func() {
				wg.Add(1)
				writeuns()
				wg.Done()
			}()
			go func() {
				wg.Add(1)
				readuns()
				wg.Done()
			}()
			wg.Wait()
		}
	})
}
/*
-= Sample Runs on Mid 2014 MacBook Pro 15" (2.5GHz Intel Core i7, 16GB 1600 MHz DDR3 RAM) =-
[1000:10 Concurrent Read:Write]$ for i in {1..5}; do go test -bench . -cpu 8 2>/dev/null; done
PASS
BenchmarkMutex-8	  300000	      3760 ns/op
BenchmarkAtomic-8	 1000000	      1031 ns/op
BenchmarkUnsafe-8	 3000000	       435 ns/op
ok  	mutexvsatomic	4.208s
PASS
BenchmarkMutex-8	  500000	      3505 ns/op
BenchmarkAtomic-8	 2000000	      1026 ns/op
BenchmarkUnsafe-8	 5000000	       462 ns/op
ok  	mutexvsatomic	7.277s
PASS
BenchmarkMutex-8	  500000	      3716 ns/op
BenchmarkAtomic-8	 1000000	      1010 ns/op
BenchmarkUnsafe-8	 5000000	       455 ns/op
ok  	mutexvsatomic	5.792s
PASS
BenchmarkMutex-8	  500000	      4057 ns/op
BenchmarkAtomic-8	 1000000	      1080 ns/op
BenchmarkUnsafe-8	 5000000	       386 ns/op
ok  	mutexvsatomic	5.696s
PASS
BenchmarkMutex-8	  500000	      3647 ns/op
BenchmarkAtomic-8	 2000000	      1026 ns/op
BenchmarkUnsafe-8	 3000000	       446 ns/op
ok  	mutexvsatomic	6.581s
[1000:100 Concurrent Read:Write]$ for i in {1..5}; do go test -bench . -cpu 8 2>/dev/null; done
PASS
BenchmarkMutex-8	  300000	      4520 ns/op
BenchmarkAtomic-8	 1000000	      4067 ns/op
BenchmarkUnsafe-8	 5000000	       424 ns/op
ok  	mutexvsatomic	8.197s
PASS
BenchmarkMutex-8	  300000	      4463 ns/op
BenchmarkAtomic-8	  300000	      4111 ns/op
BenchmarkUnsafe-8	 3000000	       440 ns/op
ok  	mutexvsatomic	4.682s
PASS
BenchmarkMutex-8	  300000	      4535 ns/op
BenchmarkAtomic-8	  500000	      3544 ns/op
BenchmarkUnsafe-8	 5000000	       428 ns/op
ok  	mutexvsatomic	5.925s
PASS
BenchmarkMutex-8	  300000	      4623 ns/op
BenchmarkAtomic-8	 1000000	      3995 ns/op
BenchmarkUnsafe-8	 5000000	       446 ns/op
ok  	mutexvsatomic	8.292s
PASS
BenchmarkMutex-8	  300000	      4471 ns/op
BenchmarkAtomic-8	  500000	      3939 ns/op
BenchmarkUnsafe-8	 5000000	       433 ns/op
ok  	mutexvsatomic	6.191s
[1000:0 Concurrent Read:Write]$ for i in {1..5}; do go test -bench . -cpu 8 2>/dev/null; done
PASS
BenchmarkMutex-8	  500000	      3599 ns/op
BenchmarkAtomic-8	 2000000	       790 ns/op
BenchmarkUnsafe-8	 3000000	       447 ns/op
ok  	mutexvsatomic	5.875s
PASS
BenchmarkMutex-8	  500000	      3553 ns/op
BenchmarkAtomic-8	 2000000	       773 ns/op
BenchmarkUnsafe-8	 3000000	       463 ns/op
ok  	mutexvsatomic	5.902s
PASS
BenchmarkMutex-8	  500000	      3660 ns/op
BenchmarkAtomic-8	 2000000	       828 ns/op
BenchmarkUnsafe-8	 3000000	       481 ns/op
ok  	mutexvsatomic	6.514s
PASS
BenchmarkMutex-8	  500000	      3748 ns/op
BenchmarkAtomic-8	 2000000	       820 ns/op
BenchmarkUnsafe-8	 3000000	       473 ns/op
ok  	mutexvsatomic	6.217s
PASS
BenchmarkMutex-8	  500000	      3724 ns/op
BenchmarkAtomic-8	 2000000	       854 ns/op
BenchmarkUnsafe-8	 3000000	       432 ns/op
ok  	mutexvsatomic	5.937s
[0:100 Concurrent Read:Write]$ for i in {1..5}; do go test -bench . -cpu 8 2>/dev/null; done
PASS
BenchmarkMutex-8	 1000000	      1378 ns/op
BenchmarkAtomic-8	 1000000	      2509 ns/op
BenchmarkUnsafe-8	 3000000	       441 ns/op
ok  	mutexvsatomic	6.167s
PASS
BenchmarkMutex-8	 1000000	      1438 ns/op
BenchmarkAtomic-8	  500000	      2601 ns/op
BenchmarkUnsafe-8	 3000000	       385 ns/op
ok  	mutexvsatomic	4.739s
PASS
BenchmarkMutex-8	 1000000	      1586 ns/op
BenchmarkAtomic-8	 1000000	      2164 ns/op
BenchmarkUnsafe-8	 3000000	       414 ns/op
ok  	mutexvsatomic	5.557s
PASS
BenchmarkMutex-8	 1000000	      1535 ns/op
BenchmarkAtomic-8	 1000000	      2179 ns/op
BenchmarkUnsafe-8	 3000000	       394 ns/op
ok  	mutexvsatomic	5.657s
PASS
BenchmarkMutex-8	 1000000	      1540 ns/op
BenchmarkAtomic-8	  500000	      2591 ns/op
BenchmarkUnsafe-8	 5000000	       351 ns/op
ok  	mutexvsatomic	5.303s
*/