最近有朋友在面試的時(shí)候被問(wèn)了select 和epoll效率差的原因,和一般人一樣,大部分都會(huì)回答select是輪詢(xún)、epoll是觸發(fā)式的,所以效率高。這個(gè)答案聽(tīng)上去很完美,大致也說(shuō)出了二者的主要區(qū)別。
今天閑來(lái)無(wú)事,翻看了下內(nèi)核代碼,結(jié)合內(nèi)核代碼和大家分享下我的觀點(diǎn)。
一、連接數(shù)
我本人也曾經(jīng)在項(xiàng)目中用過(guò)select和epoll,對(duì)于select,感觸最深的是linux下select最大數(shù)目限制(windows 下似乎沒(méi)有限制),每個(gè)進(jìn)程的select最多能處理FD_SETSIZE個(gè)FD(文件句柄),
如果要處理超過(guò)1024個(gè)句柄,只能采用多進(jìn)程了。
常見(jiàn)的使用slect的多進(jìn)程模型是這樣的: 一個(gè)進(jìn)程專(zhuān)門(mén)accept,成功后將fd通過(guò)unix socket傳遞給子進(jìn)程處理,父進(jìn)程可以根據(jù)子進(jìn)程負(fù)載分派。曾經(jīng)用過(guò)1個(gè)父進(jìn)程+4個(gè)子進(jìn)程 承載了超過(guò)4000個(gè)的負(fù)載。
這種模型在我們當(dāng)時(shí)的業(yè)務(wù)運(yùn)行的非常好。epoll在連接數(shù)方面沒(méi)有限制,當(dāng)然可能需要用戶(hù)調(diào)用API重現(xiàn)設(shè)置進(jìn)程的資源限制。
二、IO差別
1、select的實(shí)現(xiàn)
這段可以結(jié)合linux內(nèi)核代碼描述了,我使用的是2.6.28,其他2.6的代碼應(yīng)該差不多吧。
先看看select:
select系統(tǒng)調(diào)用的代碼在fs/Select.c下,
asmlinkage long sys_select(int n, fd_set __user *inp, fd_set __user *outp,
fd_set __user *exp, struct timeval __user *tvp)
{
struct timespec end_time, *to = NULL;
struct timeval tv;
int ret;
if (tvp) {
if (copy_from_user(&tv, tvp, sizeof(tv)))
return -EFAULT;
to = &end_time;
if (poll_select_set_timeout(to,
tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
(tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
return -EINVAL;
}
ret = core_sys_select(n, inp, outp, exp, to);
ret = poll_select_copy_remaining(&end_time, tvp, 1, ret);
return ret;
}
前面是從用戶(hù)控件拷貝各個(gè)fd_set到內(nèi)核空間,接下來(lái)的具體工作在core_sys_select中,
core_sys_select->do_select,真正的核心內(nèi)容在do_select里:
int do_select(int n, fd_set_bits *fds, struct timespec *end_time)
{
ktime_t expire, *to = NULL;
struct poll_wqueues table;
poll_table *wait;
int retval, i, timed_out = 0;
unsigned long slack = 0;
rcu_read_lock();
retval = max_select_fd(n, fds);
rcu_read_unlock();
if (retval < 0)
return retval;
n = retval;
poll_initwait(&table);
wait = &table.pt;
if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
wait = NULL;
timed_out = 1;
}
if (end_time && !timed_out)
slack = estimate_accuracy(end_time);
retval = 0;
for (;;) {
unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp;
set_current_state(TASK_INTERRUPTIBLE);
inp = fds->in; outp = fds->out; exp = fds->ex;
rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex;
for (i = 0; i < n; ++rinp, ++routp, ++rexp) {
unsigned long in, out, ex, all_bits, bit = 1, mask, j;
unsigned long res_in = 0, res_out = 0, res_ex = 0;
const struct file_operations *f_op = NULL;
struct file *file = NULL;
in = *inp++; out = *outp++; ex = *exp++;
all_bits = in | out | ex;
if (all_bits == 0) {
i += __NFDBITS;
continue;
}
for (j = 0; j < __NFDBITS; ++j, ++i, bit <<= 1) {
int fput_needed;
if (i >= n)
break;
if (!(bit & all_bits))
continue;
file = fget_light(i, &fput_needed);
if (file) {
f_op = file->f_op;
mask = DEFAULT_POLLMASK;
if (f_op && f_op->poll)
mask = (*f_op->poll)(file, retval ? NULL : wait);
fput_light(file, fput_needed);
if ((mask & POLLIN_SET) && (in & bit)) {
res_in |= bit;
retval++;
}
if ((mask & POLLOUT_SET) && (out & bit)) {
res_out |= bit;
retval++;
}
if ((mask & POLLEX_SET) && (ex & bit)) {
res_ex |= bit;
retval++;
}
}
}
if (res_in)
*rinp = res_in;
if (res_out)
*routp = res_out;
if (res_ex)
*rexp = res_ex;
cond_resched();
}
wait = NULL;
if (retval || timed_out || signal_pending(current))
break;
if (table.error) {
retval = table.error;
break;
}
/*
* If this is the first loop and we have a timeout
* given, then we convert to ktime_t and set the to
* pointer to the expiry value.
*/
if (end_time && !to) {
expire = timespec_to_ktime(*end_time);
to = &expire;
}
if (!schedule_hrtimeout_range(to, slack, HRTIMER_MODE_ABS))
timed_out = 1;
}
__set_current_state(TASK_RUNNING);
poll_freewait(&table);
return retval;
}
上面的代碼很多,其實(shí)真正關(guān)鍵的代碼是這一句:
mask = (*f_op->poll)(file, retval ? NULL : wait);
這個(gè)是調(diào)用文件系統(tǒng)的 poll函數(shù),不同的文件系統(tǒng)poll函數(shù)自然不同,由于我們這里關(guān)注的是tcp連接,而socketfs的注冊(cè)在 net/Socket.c里。
register_filesystem(&sock_fs_type);
socket文件系統(tǒng)的函數(shù)也是在net/Socket.c里:
static const struct file_operations socket_file_ops = {
.owner = THIS_MODULE,
.llseek = no_llseek,
.aio_read = sock_aio_read,
.aio_write = sock_aio_write,
.poll = sock_poll,
.unlocked_ioctl = sock_ioctl,
#ifdef CONFIG_COMPAT
.compat_ioctl = compat_sock_ioctl,
#endif
.mmap = sock_mmap,
.open = sock_no_open, /* special open code to disallow open via /proc */
.release = sock_close,
.fasync = sock_fasync,
.sendpage = sock_sendpage,
.splice_write = generic_splice_sendpage,
.splice_read = sock_splice_read,
};
從sock_poll跟隨下去,
最后可以到 net/ipv4/tcp.c的
unsigned int tcp_poll(struct file *file, struct socket *sock, poll_table *wait)
這個(gè)是最終的查詢(xún)函數(shù),
也就是說(shuō)select 的核心功能是調(diào)用tcp文件系統(tǒng)的poll函數(shù),不停的查詢(xún),如果沒(méi)有想要的數(shù)據(jù),主動(dòng)執(zhí)行一次調(diào)度(防止一直占用cpu),直到有一個(gè)連接有想要的消息為止。
從這里可以看出select的執(zhí)行方式基本就是不同的調(diào)用poll,直到有需要的消息為止,如果select 處理的socket很多,這其實(shí)對(duì)整個(gè)機(jī)器的性能也是一個(gè)消耗。
2、epoll的實(shí)現(xiàn)
epoll的實(shí)現(xiàn)代碼在 fs/EventPoll.c下,
由于epoll涉及到幾個(gè)系統(tǒng)調(diào)用,這里不逐個(gè)分析了,僅僅分析幾個(gè)關(guān)鍵點(diǎn),
第一個(gè)關(guān)鍵點(diǎn)在
static int ep_insert(struct eventpoll *ep, struct epoll_event *event,
struct file *tfile, int fd)
這是在我們調(diào)用sys_epoll_ctl 添加一個(gè)被管理socket的時(shí)候調(diào)用的函數(shù),關(guān)鍵的幾行如下:
epq.epi = epi;
init_poll_funcptr(&epq.pt, ep_ptable_queue_proc);
/*
* Attach the item to the poll hooks and get current event bits.
* We can safely use the file* here because its usage count has
* been increased by the caller of this function. Note that after
* this operation completes, the poll callback can start hitting
* the new item.
*/
revents = tfile->f_op->poll(tfile, &epq.pt);
這里也是調(diào)用文件系統(tǒng)的poll函數(shù),不過(guò)這次初始化了一個(gè)結(jié)構(gòu),這個(gè)結(jié)構(gòu)會(huì)帶有一個(gè)poll函數(shù)的callback函數(shù):ep_ptable_queue_proc,
在調(diào)用poll函數(shù)的時(shí)候,會(huì)執(zhí)行這個(gè)callback,這個(gè)callback的功能就是將當(dāng)前進(jìn)程添加到 socket的等待進(jìn)程上。
static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead,
poll_table *pt)
{
struct epitem *epi = ep_item_from_epqueue(pt);
struct eppoll_entry *pwq;
if (epi->nwait >= 0 && (pwq = kmem_cache_alloc(pwq_cache, GFP_KERNEL))) {
init_waitqueue_func_entry(&pwq->wait, ep_poll_callback);
pwq->whead = whead;
pwq->base = epi;
add_wait_queue(whead, &pwq->wait);
list_add_tail(&pwq->llink, &epi->pwqlist);
epi->nwait++;
} else {
/* We have to signal that an error occurred */
epi->nwait = -1;
}
}
注意到參數(shù) whead
實(shí)際上是 sk->sleep,其實(shí)就是將當(dāng)前進(jìn)程添加到sk的等待隊(duì)列里,當(dāng)該socket收到數(shù)據(jù)或者其他事件觸發(fā)時(shí),會(huì)調(diào)用
sock_def_readable
或者sock_def_write_space
通知函數(shù)來(lái)喚醒等待進(jìn)程,這2個(gè)函數(shù)都是在socket創(chuàng)建的時(shí)候填充在sk結(jié)構(gòu)里的。
從前面的分析來(lái)看,epoll確實(shí)是比select聰明的多、輕松的多,不用再苦哈哈的去輪詢(xún)了。
posted on 2010-07-10 18:40
feixuwu 閱讀(10247)
評(píng)論(3) 編輯 收藏 引用 所屬分類(lèi):
游戲開(kāi)發(fā)