thomlinton/freecache

#FreeCache - A cache library for Go with zero GC overhead.

Long lived objects in memory introduce expensive GC overhead, the GC latency can go up to hundreds of milliseconds with just a few millions of live objects. With FreeCache, you can cache unlimited number of objects in memory without increased GC latency.

##About GC Pause Issue

Here is the demo code for the GC pause issue, you can run it yourself. On my laptop, GC pause with FreeCache is under 200us, but with map, it is more than 300ms.

package main

import (
    "fmt"
    "github.com/coocood/freecache"
    "runtime"
    "runtime/debug"
    "time"
)

var mapCache map[string][]byte

func GCPause() time.Duration {
    runtime.GC()
    var stats debug.GCStats
    debug.ReadGCStats(&stats)
    return stats.Pause[0]
}

func main() {
    n := 3000 * 1000
    freeCache := freecache.NewCache(512 * 1024 * 1024)
    debug.SetGCPercent(10)
    for i := 0; i < n; i++ {
        key := fmt.Sprintf("key%v", i)
        val := make([]byte, 10)
        freeCache.Set([]byte(key), val, 0)
    }
    fmt.Println("GC pause with free cache:", GCPause())
    freeCache = nil
    mapCache = make(map[string][]byte)
    for i := 0; i < n; i++ {
        key := fmt.Sprintf("key%v", i)
        val := make([]byte, 10)
        mapCache[key] = val
    }
    fmt.Println("GC pause with map cache:", GCPause())
}

##Features

Store hundreds of millions of entries
Zero GC overhead
High concurrent thread-safe access
Pure Go implementation
Expiration support
Nearly LRU algorithm
Strictly limited memory usage
Come with a toy server that supports a few basic Redis commands with pipeline

##Performance Here is the benchmark result compares to built-in map, Set performance is about 2x faster than built-in map, Get performance is about 1/2x slower than built-in map. Since it is single threaded benchmark, in multi-threaded environment, FreeCache should be many times faster than single lock protected built-in map.

BenchmarkCacheSet        3000000               446 ns/op
BenchmarkMapSet          2000000               861 ns/op
BenchmarkCacheGet        3000000               517 ns/op
BenchmarkMapGet         10000000               212 ns/op

##Example Usage

cacheSize := 100 * 1024 * 1024
cache := freecache.NewCache(cacheSize)
debug.SetGCPercent(20)
key := []byte("abc")
val := []byte("def")
expire := 60 // expire in 60 seconds
cache.Set(key, val, expire)
got, err := cache.Get(key)
if err != nil {
    fmt.Println(err)
} else {
    fmt.Println(string(got))
}
affected := cache.Del(key)
fmt.Println("deleted key ", affected)
fmt.Println("entry count ", cache.EntryCount())

##Notice

Recommended Go version is 1.4.
Memory is preallocated.
If you allocate large amount of memory, you may need to set debug.SetGCPercent() to a much lower percentage to get a normal GC frequency.

##How it is done FreeCache avoids GC overhead by reducing the number of pointers. No matter how many entries stored in it, there are only 512 pointers. The data set is sharded into 256 segments by the hash value of the key. Each segment has only two pointers, one is the ring buffer that stores keys and values, the other one is the index slice which used to lookup for an entry. Each segment has its own lock, so it supports high concurrent access.

##TODO

Support dump to file and load from file.
Support resize cache size at runtime.

##License The MIT License

thomlinton / freecache

About

Languages