Benchmarking Go Synchronization Primitives

This repository contains some simple benchmarks of some different synchronization primitives in Golang.

Run with

$ go test -bench .

The benchmarks are implemented as a loop:

for {
    takeLock()
    doWork()
    dropLock()
}

The work done is purely a simple CPU exercise with no memory allocations. Different workloads will probably change the outcome by a lot.

Results for Different Go Versions

These numbers are from an unreliable developer laptop - so take with a grain of salt. The numbers changed in a 10% range between different runs, but the overall trends remained the same.

go version go1.14.2 linux/amd64
goos: linux
goarch: amd64
BenchmarkSync/Atomic-8         	100000000	        11.2 ns/op
BenchmarkSync/ch0-8            	100000000	        12.7 ns/op
BenchmarkSync/ch1-8            	85027264	        13.8 ns/op
BenchmarkSync/ch10-8           	82793593	        14.2 ns/op
BenchmarkSync/Mutex-8          	57655360	        19.6 ns/op
BenchmarkSync/RWMutex-8        	19018388	        90.8 ns/op
BenchmarkSync/Spamch0-8        	100000000	        13.6 ns/op
BenchmarkSync/Spamch1-8        	72175544	        15.9 ns/op
BenchmarkSync/Spamch10-8       	75115940	        16.2 ns/op
BenchmarkSync/SpamRWMutex-8    	10031592	       105 ns/op
PASS
ok  	_/home/kamstrup/Axiom/hax	13.499s

go version go1.13.8 linux/amd64
goos: linux
goarch: amd64
BenchmarkSync/Atomic-8         	100000000	        10.2 ns/op
BenchmarkSync/ch0-8            	92577460	        13.7 ns/op
BenchmarkSync/ch1-8            	81593481	        15.1 ns/op
BenchmarkSync/ch10-8           	78057223	        15.0 ns/op
BenchmarkSync/Mutex-8          	55968832	        20.7 ns/op
BenchmarkSync/RWMutex-8        	15184892	        79.9 ns/op
BenchmarkSync/Spamch0-8        	100000000	        14.2 ns/op
BenchmarkSync/Spamch1-8        	63595525	        16.9 ns/op
BenchmarkSync/Spamch10-8       	68215941	        17.2 ns/op
BenchmarkSync/SpamRWMutex-8    	 9844534	       104 ns/op
PASS
ok  	_/home/kamstrup/Axiom/hax	12.932s

go version go1.12.8 linux/amd64
goos: linux
goarch: amd64
BenchmarkSync/Atomic-8         	50000000	        37.5 ns/op
BenchmarkSync/ch0-8            	200000000	         8.21 ns/op
BenchmarkSync/ch1-8            	100000000	        13.0 ns/op
BenchmarkSync/ch10-8           	100000000	        14.2 ns/op
BenchmarkSync/Mutex-8          	50000000	        26.2 ns/op
BenchmarkSync/RWMutex-8        	50000000	        96.3 ns/op
BenchmarkSync/Spamch0-8        	100000000	        13.9 ns/op
BenchmarkSync/Spamch1-8        	100000000	        16.0 ns/op
BenchmarkSync/Spamch10-8       	100000000	        16.7 ns/op
BenchmarkSync/SpamRWMutex-8    	10000000	       142 ns/op
PASS
ok  	_/home/kamstrup/Axiom/hax	21.329s

Read the Code

Please read the code before drawing your own conclusions, it has been highly tailored for what we needed. There are some interesting caveats in the comments as well!