Calculation of RTO in tcp and related open source implementation (Linux kernel, unbound) --- rtt, rttvar, cwnd, ssthresh

https://titanwolf.org/Network/Articles/Article?AID=6d260d47-8ba6-4fa6-ba2f-4dc4c2dd6800#gsc.tab=0

 

Calculation of RTO in tcp and related open source implementation (Linux kernel, unbound)

1. Concept

        RTT: Round trip delay. 

        RTO: After tcp sends a data packet, it will start a retransmission timer, RTO is the retransmission time of this timer.

        Since RTO refers to the estimated timeout time of sending the current data packet this time, then RTO needs a good statistical method to better predict the timeout time of this time.

        SRTT: smoothed tound-trip time

        RTTVAR: round-trip time variation

Second, we can think of the calculation method (take the average)

        For example, the first RTT is 500 milliseconds and the second is 800 milliseconds, then the RTO should be 650 milliseconds when it is sent for the third time.

3. RFC793 calculation method

        Similar to taking the average.

3.1 Method
        SRTT = ( ALPHA * SRTT ) + ( ( 1-ALPHA ) * RTT )

        RTO = min [ UBOUND, max [ LBOUND, ( BETA * SRTT ) ] ]

3.2 Parameter meaning

        UBOUND is the maximum value of RTO;

        LBOUND is the minimum value of RTO;

        The value of ALPHA is 0.8 ~ 0.9;

        The value of BETA is 1.3 ~ 2.0.

RFC793: http://tools.ietf.org/html/rfc793

4. Calculation method of paper "Congestion Avoidance and Control"

        Compared with RFC793, mean deviation is increased.

4.1 Method

        Err ≡ m−a
        a ← a + gErr
        v ← v + g (| Err | −v)

4.2 Parameter meaning

        m is RTT;

        a is SRTT;

        v is the mean difference;

        g is a factor.

4.3 Method changes

Since there may be floating point numbers in the presence of g, we only need to make a change so that g = 1/2 ^ n, then the above equation becomes the following

        2 ^ na ← 2 ^ na + Err 
        2 ^ nv ← 2 ^ nv + g (| Err | −v)

4.4 Pseudocode

/∗ update Average estimator ∗/
m − = (sa >> 3); 
sa + = m;
/∗ update Deviation estimator ∗/
if (m <0)
   m = −m; 
m − = (sv >> 2);
sv + = m;
rto = (sa >> 3) + sv;

paper "Congestion Avoidance and Control": http://ee.lbl.gov/papers/congavoid.pdf

Five, RFC6298 calculation method

5.1 Calculation method

At the first RTT:

        SRTT <- R
        RTTVAR <- R/2
        RTO <- SRTT + max (G, K*RTTVAR)

During subsequent RTT:

        RTTVAR <- (1 - beta) * RTTVAR + beta * |SRTT - R'|
        SRTT <- (1 - alpha) * SRTT + alpha * R'
        RTO <- SRTT + max (G, K*RTTVAR)

5.2 Parameter meaning

R and R 'are RTT;

        alpha = 1/8;

        beta = 1/4;

        K = 4.

5.3 Changed calculation method (this method is the same as the calculation method above)

delta = RTT - SRTT
SRTT = SRTT + g * delta  
rttvar = rttvar + h(|delta| - rttvar) 
RTO = srtt + 4*rttvar

among them:

g = 1/8

h = 1/4

RFC6298: http://tools.ietf.org/html/rfc6298

6. Some open source implementations of computing RTO

In the Linux kernel: net/ipv4/tcp_input.c tcp_rtt_estimator ()

Unbound: util/rtt.c

Reference materials:

RFC793: http://tools.ietf.org/html/rfc793

paper "Congestion Avoidance and Control": http://ee.lbl.gov/papers/congavoid.pdf

RFC6298: http://tools.ietf.org/html/rfc6298

Calculation of RTO in tcp and implementation under linux: http://www.cnblogs.com/hfww/archive/2012/08/01/2618109.html

TCP reliable transmission and flow control series 3: Retransmission time RTO calculation: http://zenhumany.blog.163.com/blog/static/1718066332010827104042708/

posted @   张同光  阅读(154)  评论(0编辑  收藏  举报
编辑推荐:
· 基于Microsoft.Extensions.AI核心库实现RAG应用
· Linux系列:如何用heaptrack跟踪.NET程序的非托管内存泄露
· 开发者必知的日志记录最佳实践
· SQL Server 2025 AI相关能力初探
· Linux系列:如何用 C#调用 C方法造成内存泄露
阅读排行:
· 无需6万激活码!GitHub神秘组织3小时极速复刻Manus,手把手教你使用OpenManus搭建本
· Manus爆火,是硬核还是营销?
· 终于写完轮子一部分:tcp代理 了,记录一下
· 别再用vector<bool>了!Google高级工程师:这可能是STL最大的设计失误
· 单元测试从入门到精通
点击右上角即可分享
微信分享提示