RTO(Retransmission TimeOut)即重传超时时间
TCP超时与重传中一个最重要的部分是对一个给定连接的往返时间（RTT）的测量。由于网络流量的变化，这个时间会相应地发生改变，TCP需要跟踪这些变化并动态调整超时时间RTO。
RFC2988中是这样描述RTO的：
“The Transmission Control Protocol(TCP) uses a retransmission timer to ensure data
delivery in the absence of any feedback from the remote data receiver.The duration of
this timer is referred to as RTO(retransmission timeout).”
RTT(Round Trip Time)由三部分组成：链路的传播时间（propagation delay),末端系统的处理时间，路由器缓存中的排队和处理时间（queuing delay）。
其中，前两个部分的值对于一个TCP连接相对固定，路由器缓存中的排队和处理时间会随着整个网络拥塞程度的变化而变化。所以RTT的变化在一定程度上反应了网络的拥塞程度。
平均偏差
平均偏差（mean deviation）,简写为mdev。
It is the mean of the distance between each value and the mean.It gives us an idea of how spread out from the center the set of values is.
Here's the formula.
通过计算平均偏差，可以知道一组数据的波动情况。
在这里，平均偏差可以用来衡量RTT的抖动情况。
RTT测量原理
RTT的测量可以采用两种方法
（1）TCP Timestamp选项
TCP时间戳选项可以用来精确的测量RTT.
RTT=当前时间- 数据包中Timestamp选项的回显时间
这个回显时间是该数据包发出去的时间，知道了数据包的接收时间（当前时间）和发送时间（回显时间）,就可以轻松的得到RTT的一个测量值。
（2）重传队列中数据包的TCP控制块
在TCP重传队列中保存着发送而未被确认的数据包，数据包skb中的TCP控制块包含着一个变量，tcp_skb_cb->when，记录了该数据包的第一次发送时间。
RTT=当前时间 - when
有人可能会问：既然不用TCP Timestamp选项就能测量出RTT,为啥还要多此一举？
这是因为方法一比方法二的功能更加强大，它们是有区别的。
“TCP must use Karn's algorithm for taking RTT samples.That is,RTT samples MUST NOT be made using segments that were retransmitted(and thus for which it is ambiguious whether the reply was for the first instance of the packet or a later instance).The only case when TCP can safely take RTT samples from retransmitted segments is when the TCP timestamp option is employed, since the timestamp option removes the ambiguity regarding which instance of the data segment triggered the acknowledgement.”
对于重传的数据包的响应，方法1可以用它来采集一个新的RTT测量样本，而方法二则不能。因为TCP Timestamp选项可以区分这个响应是原始数据包还是重传数据包触发的，从而计算出准确的RTT值。
RTT测量实现
发送方每接收到一个ACK，都会调用tcp_ack()来处理。
tcp_ack()中会调用tcp_clean_rtx_queue()来删除重传队列中已经被确认的数据段。
在tcp_clean_rtx_queue()中：
如果ACK确认了重传的数据包，则seq_rtt=-1;
否则，seq_rtt = now - scb->when；
然后调用tcp_ack_update_rtt(sk,flag,seq_rtt)来更新RTT和RTO.
[java]
static inline void tcp_ack_update_rtt(struct sock *sk,const in flag,const s32 seq_rtt)
{
const struct tcp_sock *tp = tcp_sk(sk);
/*Note that peer MAY send zero echo.In this case it is ignored.(rfc1323)*/
/*如果有启用TCP Timestamp选项，且接收方的回显不为0*/
if(tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr)
tcp_ack_saw_tstamp(sk,flag);/*方法一*/
else if(seq_rtt >= 0) /*不能是重传数据包的ACK */
tcp_ack_no_tstamp(sk,seq_rtt,flag);/*方法二*/
}
方法一：tcp_ack_saw_tstamp()
[java]
/* Read draft-ietf-tcplw-high-performance before mucking with this code.
* (Supersedes RFC1323)
*/
staticvoidtcp_ack_saw_tstamp(struct sock *sk,intflag)
{
/* RTTM Rule : A TSecr value received in a segment is used to update the
* averaged RTT measurement only if the segment acknowledges some new
* data, i.e., only if it advances the left edge of the send window.
*
* See draft-ietf-tcplw-high-performance-00, section 3.3.
* 1998/04/10 Andrey V. Savochkin saw@msu.ru
*
* Changed : reset backoff as soon as we see the first valid sample.
* If we do not, we get strongly overestimated rto. With timestamps
* samples are accepted even from very old segments: f.e., when
* rtt = 1 increases to 8, we retransmit 5 times and after 8 seconds
* delayed answer arrives rto becomes 120 seconds! If at least one
* of segments in window is lost... Volia.
* ——ANK(010210)
*/
struct tcp_sock *tp = tcp_sk(sk);
/* RTT采样值：now - rcv_tsecr */
tcp_valid_rtt_meas(sk, tcp_time_stamp - tp->rx_opt.rcv_tsecr);
}
方法二：tcp_ack_no_tstamp()
[java]
staticvoidtcp_ack_no_tstamp(struct sock *sk, u32 seq_rtt,intflag)
{
/* We don't have a timestsamp. Can only use packets that are not
* retransmitted to determine rtt estimates. Also, we must not reset
* the backoff for rto until we get a non-retransmitted packet. This
* allows us to deal with a situation where the network delay has
* increased suddenly. I.e. Karn's algorithm. (SIGCOMM '87, p5.)
*/
if(flag & FLAG_RETRANS_DATA_ACKED)
return;/* 如果ACK确认的是重传的数据包，则不进行RTT采样*/
/* RTT采样值：seq_rtt，这个值是在tcp_clean_rtx_queue()中计算得到的。*/
tcp_valid_rtt_meas(sk, seq_rtt);
}
OK,到这边RTT的测量已经结束了，接下来就是RTO值得计算
RTO计算原理
涉及到的变量
[java]
struct tcp_sock {
...
/* RTT measurement */
u32 srtt;/* smoothed round trip time << 3 */
u32 mdev;/* medium deviation */
u32 mdev_max;/* maximal mdev for the last rtt period */
u32 rttvar/* smoothed mdev_max */
u32 rtt_seq;/* sequence number to update rttvar */
...
}
srtt为经过平滑后的RTT值，它代表着当前的RTT值，每收到一个ACK更新一次。
为了避免浮点运算，它是实际RTT值的8倍。
mdev为RTT的平均偏差，用来衡量RTT的抖动，每收到一个ACK更新一次。
mdev_max为上一个RTT内的最大mdev，代表上个RTT内时延的波动情况，有效期为一个RTT.
rttvar为mdev_max的平滑值，可升可降，代表着连接的抖动情况，在连接断开前都有效。
“To compute the current RTO，a TCP sender maintains two state variables,SRTT(smoothed round-trip time) and RTTVAR（round-trip time variation）.”
实际上，RTO = srtt>>3+rttvar.
rtt表示新的RTT测量值。
old_srtt表示srtt更新前的srtt>>3，即旧的srtt值。
new_srtt表示srtt更新后的srtt>>3,即新的srtt值。
old_mdev表示旧的mdev。
new_mdev表示更新后的mdev。
（1）获得第一个RTT测量值
srtt = rtt<<3;
mdev=rtt<<1;
mdev_max = rttvar = max(mdev,rto_min);
所以，获得第一个RTT测量值后的RTO = rtt+rttvar，如果mdev=2*rtt>rto_min,
那么RTO = 3*rtt;否则 RTO=rtt+rto_min.
（2）获得第一个RTT测量值
srtt = rtt<<3;
mdev = rtt<<1;
mdev_max=rttvar=max(mdev,rto_min);
所以获得第一个RTT测量值后的RTO=rtt+rttvar,如果mdev = 2*rtt>rto_min,
那么RTO=3*rtt;否则，RTO=rtt+rto_min。
(2)获得第n个RTT测量值（n>=2）
srtt的更新：new_srtt = 7/8 old_srtt+1/8 rtt
mdev的更新：
err=rtt-old_srtt
当RTT变小时，即err<0时
1）如果|err|>1/4 old_mdev，则new_mdev = 31/32 old_mdev + 1/8|err|
此时：old_mdev<new_mdev<3/4 old_mdev + |err|
new_mdev有稍微变大，但是增大得不多。由于RTT是变小，所以RTO也要变小，如果
new_mdev增大很多（比如：new_mdev = 3/4 old_mdev+|err|）,就会导致RTO变大，不符合我们的预期
“This is similar to one of Eifel findings.Eifel blocks mdev updates when rtt decreases.
This solution is a bit different:we use finer gain for mdev in this case(alpha *beta).
Like Eifel it also prevents growth of rto,but also it limits too fast rto decreases,happening in pure Eifel.”
2)如果|err|<=1/4 old_mdev,则new_mdev=3/4 old_mdev + |err|
此时：new_mdev < old_mdev
new_mdev变小，会导致RTO变小，符合我们的预期。
当RTT变大时，即err>0时
new_mdev = 3/4 old_mdev + |err|
此时：new_mdev > old_mdev
new_mdev变大，会导致RTO变小，这也符合我们的预期。
mdev_max和rttvar的更新
在每个RTT开始时，mdev_max = rto_min
如果在此RTO内，有更大的mdev，则更新mdev_max。
如果mdev_max > rttvar,则rttvar = mdev_max;
否则，本RTT结束后，rttvar -=(rttvar - mdev_max)>>2。
这样一来，就可以通过mdev_max来调节rttvar,间接的调节RTO。
RTO计算实现
不管是方法一还是方法二，最终都调用tcp_valid_rtt_means()来更新RTT和RTO.
[java]
/* seq_rtt为此次得到的RTT测量值。*/
voidtcp_valid_rtt_meas(struct sock *sk, u32 seq_rtt)
{
tcp_rtt_estimator(sk, seq_rtt);/* 更新相关值*/
tcp_set_rto(sk);/*设置新的RTO*/
inet_csk(sk)->icsk_backoff =0;/* 清零退避指数*/
}
RTO = srtt>>8+rttvar。而srtt和rttvar的更新都是在tcp_rtt_estimator()来进行。
[java]
/* Called to compute a smoothed rtt estimate. The data fed to this
* routine either comes from timestamps, or from segments that were
* known _not_ to have been retransmitted [see Karn/Partridge Proceedings
* SIGCOMM 87]. The algorithm is from the SIGCOMM 88 piece by Van
* Jacobson.
* NOTE : the next three routines used to be one big routine.
* To save cycles in the RFC 1323 implementation it was better to break it
* up into three procedures. ——erics
*/
staticvoidtcp_rtt_estimator (struct sock *sk,const__u32 mrtt)
{
struct tcp_sock *tp = tcp_sk(sk);
longm = mrtt;/*此为得到的新的RTT测量值*/
/* The following amusing code comes from Jacobson's article in
* SIGCOMM '88. Note that rtt and mdev are scaled versions of rtt and
* mean deviation. This is designed to be as fast as possible
* m stands for "measurement".
*
* On a 1990 paper the rto value is changed to :
* RTO = rtt + 4 * mdev
*
* Funny. This algorithm seems to be very broken.
* These formulae increase RTO, when it should be decreased, increase
* too slowly, when it should be increased quickly, decrease too quickly
* etc. I guess in BSD RTO takes ONE value, so that it is absolutely does
* not matter how to calculate it. Seems, it was trap that VJ failed to
* avoid. 8)
*/
if(m ==0)
m =1;/* RTT的采样值不能为0 */
/* 不是得到第一个RTT采样*/
if(tp->srtt !=0) {
m -= (tp->srtt >>3);/* m is now error in rtt est */
tp->srtt += m;/* rtt = 7/8 rtt + 1/8 new ，更新srtt*/
if(m <0) {/*RTT变小*/
m = -m;/* m is now abs(error) */
m -= (tp->mdev >>2);/* similar update on mdev */
/* This is similar to one of Eifel findings.
* Eifel blocks mdev updates when rtt decreases.
* This solution is a bit different : we use finer gain
* mdev in this case (alpha * beta).
* Like Eifel it also prevents growth of rto, but also it
* limits too fast rto decreases, happening in pure Eifel.
*/
if(m >0)/* |err| > 1/4 mdev */
m >>=3;
}else{/* RTT变大 */
m -= (tp->mdev >>2);/* similar update on mdev */
}
tp->mdev += m;/* mdev = 3/4 mdev + 1/4 new，更新mdev */
/* 更新mdev_max和rttvar */
if(tp->mdev > tp->mdev_max) {
tp->mdev_max = tp->mdev;
if(tp->mdev_max > tp->rttvar )
tp->rttvar = tp->mdev_max;
}
/* 过了一个RTT了，更新mdev_max和rttvar */
if(after(tp->snd_una, tp->rtt_seq)) {
if(tp->mdev_max < tp->rttvar)/*减小rttvar */
tp->rttvar -= (tp->rttvar - tp->mdev_max) >>2;
tp->rtt_seq = tp->snd_nxt;
tp->mdev_max = tcp_rto_min(sk);/*重置mdev_max */
}
if(after(tp->snd_una, tp->rtt_seq)) {
if(tp->mdev_max < tp->rttvar)/*减小rttvar */
tp->rttvar -= (tp->rttvar - tp->mdev_max) >>2;
tp->rtt_seq = tp->snd_nxt;
tp->mdev_max = tcp_rto_min(sk);/*重置mdev_max */
}
}else{
/* 获得第一个RTT采样*/
/* no previous measure. */
tp->srtt = m <<3;/* take the measured time to be rtt */
tp->mdev = m <<1;/* make sure rto = 3 * rtt */
tp->mdev_max = tp->rttvar = max(tp->mdev, tcp_rto_min(sk));
tp->rtt_seq = tp->snd_nxt;/*设置更新mdev_max的时间*/
}
}
rto_min的取值如下：
[java]
/* 最大的RTO为120s，指数退避时不能超过这个值 */
#define TCP_RTO_MAX ((unsigned) (120*HZ))
/* 最小的RTO为200ms，rttvar不能低于这个值 */
#define TCP_RTO_MIN ((unsigned) (HZ /5))
/* 还没有计算出RTO值前的RTO初始值，为1s */
#define TCP_TIMEOUT_INIT ((unsigned) (1* HZ))
/* Compute the actual rto_min value */
staticinline u32 tcp_rto_min (struct sock *sk)
{
conststruct dst_entry *dst = __sk_dst_get(sk);
u32 rto_min = TCP_RTO_MIN;
/*如果路由缓存中存在RTO_MIN，则取其为最小RTO*/
if(dst && dst_metric_locked(dst, RTAX_RTO_MIN))
rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN));
returnrto_min;
}
RTO的设置
[java]
/* Calculate rto without backoff. This is the second half of Van Jacobson's
* routine referred to above.
*/
staticinlinevoidtcp_set_rto(struct sock *sk)
{
conststruct tcp_sock *tp = tcp_sk(sk);
inet_csk(sk)->icsk_rto = __tcp_set_rto(tp);
tcp_bound_rto(sk);
}
staticinline u32 __tcp_set_rto(conststruct tcp_sock *tp)
{
return(tp->srtt >>3) + tp->rttvar;
}
staticinlinevoidtcp_bound_rto(conststruct sock *sk)
{
if(inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
}
函数调用
以上涉及到的函数调用关系如下：
总结
早期的RTT的测量是采用粗粒度的定时器（Coarse grained timer）,这会有比较大的误差。
现在由于TCP Timestamp选项的使用，能够更精确的测量RTT,从而计算出更加准确的RTO
