28

GitHub - linux4life798/safetyfast: An Go library of synchronization primitives t...

 5 years ago
source link: https://github.com/linux4life798/safetyfast
Go to the source link to view the article. You can view the picture content, updated content and better typesetting reading experience. If the link is broken, please click the button below to view the snapshot at that time.

README.md

GoDoc Go Report Card codecov Build Status

SafetyFast - Put thread-safety first, with the performance of safety last.

This is a Go library that implements synchronization primitives over Intel TSX (hardware transactional primitives).

go get github.com/linux4life798/safetyfast

Checkout the SafetyFast Project Page.

Benchmarking

The following plot shows the number of milliseconds it took for 8 goroutines to increments 480000 random elements (per goroutine) of an array of ints. The x axis denotes how large (and therefore sparse) the array was. The synchronization primitive used during the increment is indicated as a series/line.

Performance Graph

Note that, as the array size increases, the likelihood of two goroutines touching the same element at the same instance decreases. This is why we see such a dramatic increase in speed, when using either the HLE or RTM style synchronization primitive.

The SystemMutex is just sync.Mutex.

It is also worth observing that the performance started to degrade towards the very large array sizes. This is most likely due to a cache size limitation.

Snippets

Using RTM

m := map[string]int{
    "word1": 0,
}

c := NewRTMContexDefault()
c.Atomic(func() {
    // Action to be done transactionally
    m["word1"] = m["word1"] + 1
})

Using HLE

m := map[string]int{
    "word1": 0,
}

var lock safetyfast.SpinHLEMutex
lock.Lock()
// Action to be done transactionally
m["word1"] = m["word1"] + 1
lock.Unlock()

Examples

Checking for HLE and RTM CPU support

It is necessary to check that the CPU you are using support Intel RTM and/or Intel HLE instruction sets, since safetyfast does not check. This can be accomplished by using the Intel provided cpuid package, as shown below.

import (
  "github.com/intel-go/cpuid"
)

func main() {
	if !cpuid.HasExtendedFeature(cpuid.RTM) {
		panic("The CPU does not support Intel RTM")
	}

	if !cpuid.HasExtendedFeature(cpuid.HLE) {
		panic("The CPU does not support Intel HLE")
	}
}

Using RTM

package main

import (
    "fmt"
    "sync"
    "github.com/linux4life798/safetyfast"
)

func main() {
    m := map[string]int{
        "word1": 0,
        "word2": 0,
    }

    c := safetyfast.NewRTMContexDefault()
    var wg sync.WaitGroup

    wg.Add(2)
    go c.Atomic(func() {
        // Action to be done transactionally
        m["word1"] = m["word1"] + 1
        wg.Done()
    })
    go c.Atomic(func() {
        // Action to be done transactionally
        m["word1"] = m["word1"] + 1
        wg.Done()
    })
    wg.Wait()

    fmt.Println("word1 =", m["word1"])
}

Using HLE

package main

import (
    "fmt"
    "sync"
    "github.com/linux4life798/safetyfast"
)

func main() {
    m := map[string]int{
        "word1": 0,
        "word2": 0,
    }

    var lock safetyfast.SpinHLEMutex
    var wg sync.WaitGroup

    wg.Add(2)
    go func() {
        lock.Lock()
        // Action to be done transactionally
        m["word1"] = m["word1"] + 1
        lock.Unlock()
        wg.Done()
    }()
    go func() {
        lock.Lock()
        // Action to be done transactionally
        m["word1"] = m["word1"] + 1
        lock.Unlock()
        wg.Done()
    }()
    wg.Wait()

    fmt.Println("word1 =", m["word1"])
}

report

Recommend

About Joyk


Aggregate valuable and interesting links.
Joyk means Joy of geeK