本文对比了poll与select机制，介绍了poll的优点如文件描述符数量不受限及输入输出参数分离，并指出了其性能瓶颈。重点讲解了epoll在Linux 2.6中的高效性，包括底层回调机制、红黑树维护、就绪队列快速访问等，并提供了基于LT与ET模式下的epoll示例代码。
poll
相对于select来说，poll的优点：
1）能监视的文件描述符的数量没有限制
2）poll输入和输出参数不是同一变量，输入和输出分离
缺点：
1）和select一样，监视的文件描述符过多时，后序也许需要进行遍历式访问，性能也会下降。（为了解决这个问题所以用到了epoll）
makefile
mypoll:mypoll.c
gcc -o $@ $^
.PHONY:clean
clean:
rm -f mypoll
mypoll
#include <stdio.h>
#include <poll.h>
int main()
{
struct pollfd ev;
ev.fd = 0;
ev.events = POLLIN;
ev.revents = 0;
int timeout = -1;
while(1)
{
switch(poll(&ev,1,timeout))
{
case 0:
sleep(1);
printf("timeout...\n");
break;
case -1:
perror("poll");
break;
default:
{
char buf[1024];
if(ev.fd == 0&& ev.revents & POLLIN)
{
ssize_t s = read(0,buf,sizeof(buf)-1);
if(s>0)
{
buf[s-1] = 0;
printf("echo %s\n",buf);
}
}
}
break;
}
}
return 0;
}
epoll
linux2.6下epoll是多路转接性能最好的
高效的原因：
1.底层采用回调机制来激活所关心的结点
2.维护所关心的文件描述符以及对应事件采用红黑树，进而可以减少维护该数据结构增删查改的成本
3.事件一旦就绪，相干事件发送到就绪队列，当上层epoll _wait在获取时可以以O1时间复杂度获取有效事件。
4.就绪队列用户到内核区采用内存映射机制，减少数据拷贝次数
5.就绪时间的陈列方式是从0下标开始连续陈列，避免了数据重复拷贝的问题
eopll的三个接口：
epoll_create 调用它会在底层创建一颗空的红黑树（只有根，没有结点）并维护就绪队列
epoll_ctl 可以向红黑树中添加或释放结点
epoll_wait 返回要关心的事件中哪些已经就绪（底层采用回调机制激活结点（激活方式有两种：LT水平触发（总是通知）和ET边沿触发（只通知一次））并放在就绪队列（队列是有顺序的，从左往右）中）
makefile
myepoll:myepoll.c
gcc -o $@ $^
.PHONY:clean
clean:
rm -f myepoll
基于LT模式下的epoll
myepoll
#include <stdio.h>
#include <sys/epoll.h>
#include <sys/types.h>
#include <netinet/in.h>
#include <arpa/inet.h>
#include <stdlib.h>
#include <string.h>
#include <sys/epoll.h>
#define SIZE 64
const char *msg ="HTTP/1.0 200 OK\r\n\r\n<html><h1>chao ge!</h1></html>\r\n";
static void usage(const char *proc)
{
printf("Usage:\n\t%s [local_ip] [local_port]\n\n",proc);
}
int startup(const char *ip, int port)
{
int sock = socket(AF_INET,SOCK_STREAM,0);
if(sock < 0)
{
perror("socket");
exit(2);
}
int opt = 1;
setsockopt(sock,SOL_SOCKET,SO_REUSEADDR,&opt,sizeof(opt));//设置端口复用
struct sockaddr_in local;
local.sin_family =AF_INET;
local.sin_port = htons(port);
local.sin_addr.s_addr = inet_addr(ip);
if(bind(sock,(struct sockaddr*)&local,sizeof(local)) < 0)
{
perror("bind");
exit(3);
}
if(listen(sock, 10))
{
perror("listen");
exit(4);
}
return sock;
}
int main(int argc,char *argv[])
{
if(argc != 3)
{
usage(argv[0]);
return 1;
}
int listen_sock = startup(argv[1],atoi(argv[2]));
int epfd = epoll_create(256);//创建模型
if(epfd < 0)
{
perror("epoll_create");
return 5;
}
printf("listen_sock: %d, epfd: %d\n",listen_sock,epfd);
struct epoll_event ev;
ev.events =EPOLLIN;//LT
ev.data.fd = listen_sock;
epoll_ctl(epfd,EPOLL_CTL_ADD,listen_sock,&ev);
int nums = -1;
int timeout =-1;
struct epoll_event revs[SIZE];//返回的事件集
while(1)
{
switch((nums = epoll_wait(epfd,revs,SIZE,timeout)))
{
case 0:
printf("timeout...\n");
break;
case -1:
perror("epoll_wait");
break;
default:
{
//at least one fd ready!
int i =0;
for(;i < nums;i++)
{
int fd =revs[i].data.fd;
if(fd == listen_sock &&(revs[i].events &EPOLLIN))
{
//listen_sock ready!
struct sockaddr_in client;
socklen_t len =sizeof(client);
int rw_sock = accept(listen_sock,(struct sockaddr*)&client,&len);
if(rw_sock < 0)
{
perror("accept");
continue;
}
printf("get a new client [%s:%d]\n",inet_ntoa(client.sin_addr),ntohs(client.sin_port));
ev.events = EPOLLIN;
ev.data.fd = rw_sock;
epoll_ctl(epfd,EPOLL_CTL_ADD,rw_sock,&ev);
}
else if(fd!= listen_sock)
{
//normal fd rw events ready!
if(revs[i].events &EPOLLIN)
{
//read
char buf[4096];
ssize_t s = read(fd,buf,sizeof(buf)-1);
if(s>0)
{
buf[s] = 0;
printf("client# %s\n",buf);
ev.events =EPOLLOUT;
ev.data.fd = fd;
epoll_ctl(epfd,EPOLL_CTL_MOD,fd,&ev);
}
else if(s == 0)
{
printf("client is quit!\n");
close(fd);
epoll_ctl(epfd,EPOLL_CTL_DEL,fd,NULL);
}
else
{
perror("read");
close(fd);
epoll_ctl(epfd,EPOLL_CTL_DEL,fd,NULL);
}
}
else if(revs[i].events &EPOLLOUT)
{
//write
write(fd,msg,strlen(msg));
close(fd);
epoll_ctl(epfd,EPOLL_CTL_DEL\
,fd,NULL);
}
else
{
}
}
else
{
}
}
}
break;
}
}
close（epfd）;
return 0;
}
基于ET模式下的epoll
#include <stdio.h>
#include <sys/epoll.h>
#include <sys/types.h>
#include <netinet/in.h>
#include <arpa/inet.h>
#include <stdlib.h>
#include <string.h>
#include <sys/epoll.h>
#include <fcntl.h>
#define SIZE 64
typedef struct fd_buff//为每一个文件描述符设置一个缓冲区
{
int fd;
char buf[4096];
int step;
}fd_buff_t,*fd_buff_p,**fd__buff__pp;
fd_buff_p alloc_buff(int sock)
{
fd_buff_p tmp =(fd_buff_p)malloc(sizeof(fd_buff_t));
if(!tmp)
{
return NULL;
}
tmp->fd = sock;
tmp->step = 0;
}
void delete_buff(fd_buff_p fp)
{
if(fp)
{
free(fp);
}
}
const char *msg ="HTTP/1.0 200 OK\r\n\r\n<html><h1>chao ge!</h1></html>\r\n";
static void usage(const char *proc)
{
printf("Usage:\n\t%s [local_ip] [local_port]\n\n",proc);
}
int startup(const char *ip, int port)
{
int sock = socket(AF_INET,SOCK_STREAM,0);
if(sock < 0)
{
perror("socket");
exit(2);
}
int opt = 1;
setsockopt(sock,SOL_SOCKET,SO_REUSEADDR,&opt,sizeof(opt));
struct sockaddr_in local;
local.sin_family =AF_INET;
local.sin_port = htons(port);
local.sin_addr.s_addr = inet_addr(ip);
if(bind(sock,(struct sockaddr*)&local,sizeof(local)) < 0)
{
perror("bind");
exit(3);
}
if(listen(sock, 10))
{
perror("listen");
exit(4);
}
return sock;
}
int set_nonblock(int fd)
{
int fl = fcntl(fd,F_GETFL);//获得文件描述符的标志位信息（确定它是阻塞还是非阻塞）
fcntl(fd,F_SETFL,fl|O_NONBLOCK);//再调用fcntl通过F_SETFL设置其描述符，为其添加一个选项：O_NONBLOCK(非阻塞)
}
int myread(int fd,char *buf,int size)
{
//循环读（一次读不完）
ssize_t len = 0;//当前长度
ssize_t total = 0;//总共读的数据
while((len = read(fd,buf+total,8))> 0&&len ==8)
{
total +=len;
if(len < 8)
{
break;
}
}
}
int main(int argc,char *argv[])
{
if(argc != 3)
{
usage(argv[0]);
return 1;
}
int listen_sock = startup(argv[1],atoi(argv[2]));
set_nonblock(listen_sock);//设置套接字为非阻塞
int epfd = epoll_create(256);
if(epfd < 0)
{
perror("epoll_create");
return 5;
}
printf("listen_sock: %d, epfd: %d\n",listen_sock,epfd);
struct epoll_event ev;
ev.events =EPOLLIN|EPOLLET;// |epollet设置为ET模式（默认为LT）
ev.data.ptr = alloc_buff(listen_sock);
epoll_ctl(epfd,EPOLL_CTL_ADD,listen_sock,&ev);
int nums = -1;
int timeout =-1;
struct epoll_event revs[SIZE];
while(1)
{
switch((nums = epoll_wait(epfd,revs,SIZE,timeout)))
{
case 0:
printf("timeout...\n");
break;
case -1:
perror("epoll_wait");
break;
default:
{
//at least one fd ready!
int i =0;
for(;i < nums;i++)
{
int fd =((fd_buff_p)(revs[i].data.ptr))->fd;
if(fd == listen_sock &&(revs[i].events &EPOLLIN))
{
//listen_sock ready!
struct sockaddr_in client;
socklen_t len =sizeof(client);
int rw_sock = accept(listen_sock,(struct sockaddr*)&client,&len);
if(rw_sock < 0)
{
perror("accept");
continue;
}
printf("get a new client [%s:%d]\n",inet_ntoa(client.sin_addr),ntohs(client.sin_port));
set_nonblock(rw_sock);
ev.events = EPOLLIN|EPOLLET;
ev.data.ptr = alloc_buff(rw_sock);
epoll_ctl(epfd,EPOLL_CTL_ADD,rw_sock,&ev);
}
else if(fd!= listen_sock)
{
//normal fd rw events ready!
if(revs[i].events &EPOLLIN)
{
//read
char buf[4096];
ssize_t s = myread(fd,buf,sizeof(buf)-1);
if(s>0)
{
buf[s] = 0;
printf("client# %s\n",buf);
ev.events =EPOLLOUT|EPOLLET;
ev.data.ptr = revs[i].data.ptr;
epoll_ctl(epfd,EPOLL_CTL_MOD,fd,&ev);
}
else if(s == 0)
{
printf("client is quit!\n");
close(fd);
epoll_ctl(epfd,EPOLL_CTL_DEL,fd,NULL);
delete_buff(revs[i].data.ptr);
}
else
{
perror("read");
close(fd);
epoll_ctl(epfd,EPOLL_CTL_DEL,fd,NULL);
}
}
else if(revs[i].events &EPOLLOUT)
{
//write
write(fd,msg,strlen(msg));
close(fd);
epoll_ctl(epfd,EPOLL_CTL_DEL,fd,NULL);
}
else
{
}
}
else
{
}
}
}
break;
}
}
return 0;
}
扫一扫
775
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
