gnet 单核和多核配置的性能问题

[charlesgreat]

多核linux和mac下配置NumEventLoop多了反而没有没有配置1的性能高，然后用pprof查看线程不管配置1还是4，都是只有一个线程（多个协程）。
怎样能让gnet确实启动多个线程呢？

[panjf2000]

你是怎么看线程数量的？具体步骤贴一下？什么输出？

[panjf2000]

我用 GODEBUG=schedtrace=1000 进行调度器追踪之后发现是有启动多个 OS 线程的，我想应该是你的设置问题吧。

SCHED 0ms: gomaxprocs=4 idleprocs=1 threads=5 spinningthreads=1 idlethreads=0 runqueue=0 [1 0 0 0]
# github.com/panjf2000/gnet
SCHED 0ms: gomaxprocs=4 idleprocs=1 threads=4 spinningthreads=1 idlethreads=0 runqueue=0 [1 0 0 0]
# command-line-arguments
SCHED 0ms: gomaxprocs=4 idleprocs=1 threads=4 spinningthreads=1 idlethreads=0 runqueue=0 [1 0 0 0]
# command-line-arguments
SCHED 0ms: gomaxprocs=4 idleprocs=2 threads=3 spinningthreads=1 idlethreads=0 runqueue=0 [0 0 0 0]
SCHED 1002ms: gomaxprocs=4 idleprocs=2 threads=12 spinningthreads=1 idlethreads=5 runqueue=0 [0 0 0 0]
SCHED 0ms: gomaxprocs=4 idleprocs=2 threads=4 spinningthreads=1 idlethreads=0 runqueue=0 [0 0 0 0]
2020/05/08 12:20:14 HTTP server is listening on [::]:8080 (multi-cores: false, loops: 4)
SCHED 2012ms: gomaxprocs=4 idleprocs=4 threads=12 spinningthreads=0 idlethreads=5 runqueue=0 [0 0 0 0]
SCHED 1001ms: gomaxprocs=4 idleprocs=4 threads=11 spinningthreads=0 idlethreads=1 runqueue=0 [0 0 0 0]
SCHED 3014ms: gomaxprocs=4 idleprocs=4 threads=12 spinningthreads=0 idlethreads=5 runqueue=0 [0 0 0 0]
SCHED 2002ms: gomaxprocs=4 idleprocs=4 threads=11 spinningthreads=0 idlethreads=1 runqueue=0 [0 0 0 0]
SCHED 4018ms: gomaxprocs=4 idleprocs=4 threads=12 spinningthreads=0 idlethreads=5 runqueue=0 [0 0 0 0]
SCHED 3002ms: gomaxprocs=4 idleprocs=4 threads=11 spinningthreads=0 idlethreads=1 runqueue=0 [0 0 0 0]
SCHED 5021ms: gomaxprocs=4 idleprocs=4 threads=12 spinningthreads=0 idlethreads=5 runqueue=0 [0 0 0 0]
SCHED 4014ms: gomaxprocs=4 idleprocs=4 threads=11 spinningthreads=0 idlethreads=1 runqueue=0 [0 0 0 0]
SCHED 6027ms: gomaxprocs=4 idleprocs=4 threads=12 spinningthreads=0 idlethreads=5 runqueue=0 [0 0 0 0]
SCHED 5019ms: gomaxprocs=4 idleprocs=4 threads=11 spinningthreads=0 idlethreads=1 runqueue=0 [0 0 0 0]
SCHED 7036ms: gomaxprocs=4 idleprocs=4 threads=12 spinningthreads=0 idlethreads=5 runqueue=0 [0 0 0 0]
SCHED 6023ms: gomaxprocs=4 idleprocs=4 threads=11 spinningthreads=0 idlethreads=1 runqueue=0 [0 0 0 0]
SCHED 8042ms: gomaxprocs=4 idleprocs=4 threads=12 spinningthreads=0 idlethreads=5 runqueue=0 [0 0 0 0]
SCHED 7023ms: gomaxprocs=4 idleprocs=4 threads=11 spinningthreads=0 idlethreads=1 runqueue=0 [0 0 0 0]
SCHED 9045ms: gomaxprocs=4 idleprocs=4 threads=12 spinningthreads=0 idlethreads=5 runqueue=0 [0 0 0 0]
SCHED 8033ms: gomaxprocs=4 idleprocs=4 threads=11 spinningthreads=0 idlethreads=1 runqueue=0 [0 0 0 0]
SCHED 10055ms: gomaxprocs=4 idleprocs=4 threads=12 spinningthreads=0 idlethreads=5 runqueue=0 [0 0 0 0]
SCHED 9038ms: gomaxprocs=4 idleprocs=4 threads=11 spinningthreads=0 idlethreads=1 runqueue=0 [0 0 0 0]
SCHED 11059ms: gomaxprocs=4 idleprocs=4 threads=12 spinningthreads=0 idlethreads=5 runqueue=0 [0 0 0 0]

[charlesgreat]

我是用如下方式测试的， 不知道为啥loops多了之后，TPS反而慢了
ab -c 100 -n 1000000 -k http://127.0.0.1:50007/index.html

(multi-cores: true, loops: 1) 的 TPS大概在200K
(multi-cores: true, loops: 2) 的 TPS大概是175K

[panjf2000]

贴一下 server 的代码我看下？

[charlesgreat]

package main

import (
	"bytes"
	"flag"
	"fmt"
	"github.com/panjf2000/gnet"
	"log"
	"net/http"
)


type httpServer struct {
	*gnet.EventServer
}

var rtn_close="HTTP/1.1 200 OK\r\nConnection: close\r\nContent-Type: text/html;charset=utf-8\r\nContent-Length:2\r\n\r\nok "
var rtn="HTTP/1.1 200 OK\r\nConnection: keep-alive\r\nContent-Type: text/html;charset=utf-8\r\nContent-Length:2\r\n\r\nok "


func (hs *httpServer) React(frame []byte, c gnet.Conn) (out []byte, action gnet.Action) {
	if c.Context() != nil {
		// bad thing happened
		action = gnet.Close
		return
	}

	if bytes.Contains(frame, []byte("Keep-Alive")) || bytes.Contains(frame, []byte("keep-alive")) {
		out = append(out, rtn...)
	} else {
		out = append(out, rtn_close...)
		action = gnet.Close
	}

	return
}

func (hs *httpServer) OnInitComplete(srv gnet.Server) (action gnet.Action) {
	log.Printf("HTTP server is listening on %s (multi-cores: %t, loops: %d)\n",
		srv.Addr.String(), srv.Multicore, srv.NumEventLoop)
	return
}

func main() {

	var port int
	var NumEventLoop int

	// Example command: go run http.go --port 8080
	flag.IntVar(&port, "port", 50007, "server port")
	flag.IntVar(&NumEventLoop, "NumEventLoop", 4, "num loops")
	flag.Parse()

	http := new(httpServer)

	// Start serving!
	log.Fatal(gnet.Serve(http, fmt.Sprintf("tcp://:%d", port), gnet.WithNumEventLoop(NumEventLoop), gnet.WithMulticore(true)))
}

[panjf2000]

抱歉，回复晚了。
首先，NumEventLoop 和 Multicore 不用两个都设置，前者会覆盖后者的。不过单看你这个代码，我看不出什么问题，你有没有试过继续增加 NumEventLoop ? 增加后性能有没有提升？

@charlesgreat

[guonaihong]

ab 是单核心测试工具，试下wrk呢。我猜和客户端有关系，客户端只能使用一个物理核心的算力，服务端加两个核心，benchmark数据也是上不去的。

[charlesgreat]

ab 是单核心测试工具，试下wrk呢。我猜和客户端有关系，客户端只能使用一个物理核心的算力，服务端加两个核心，benchmark数据也是上不去的。

上不去也不应该下降吧

[panjf2000]

可能是测试结果波动？多测几次也是这样吗？或者试试只设置 gnet.WithMulticore(true) 呢？

[cctse]

通过下面设置单核真的比默认开启多核性能高
runtime.GOMAXPROCS(1)

@panjf2000

[charlesgreat]

测试代码业务处理很快，瓶颈在I/0， 启动多核后也无法解决I/0的瓶颈
如果业务代码加点耗时的话，配置多核会比单核快

[ballontt]

测试代码业务处理很快，瓶颈在I/0， 启动多核后也无法解决I/0的瓶颈
如果业务代码加点耗时的话，配置多核会比单核快

请问这个问题得到了验证吗

[lomoye]

我用wrk本地单机测试，同样有这个问题，服务端代码跑的是官网上提供的那个http的例子。
gnet.WithMulticore(multicore)这个multicore设置为false并发可以达到90000左右，设为true只有80000多一点。。

开启多核后：

[someview]

我们利用gnet实现的ws框架，性能4核的时候是1核的2倍多一点，很好奇这个cpu 亲和性是怎么做的

[github-actions]

This issue is marked as stale because it has been open for 30 days with no activity.

You should take one of the following actions:

• Manually close this issue if it is no longer relevant
• Comment if you have more information to share

This issue will be automatically closed in 7 days if no further activity occurs.

[github-actions]

This issue was closed because it has been inactive for 7 days since being marked as stale.

If you believe this is a false alarm, please leave a comment for it or open a new issue, you can also reopen this issue directly if you have permission.