1. TCP Timeout and Retransmission
Chap 1 Foundation
TCP Timeout and Retransmission
(월)
이 병 희

2. Introduction TCP provides a reliable transport
Each end send acknowledgment to the other end when receive data
Data and acknowledgment can get lost
TCP handles this by setting a timeout when it sends data and retransmits data
TCP manages four different timers
retransmission timer
persist timer
keepalive timer
2MSL timer

3. RTT Measurement Timeout should be based on Round-trip time (RTT)
Chap 1 Foundation
RTT Measurement
Timeout should be based on Round-trip time (RTT)
RTT measured by observing difference between time of transmission and arrival of acknowledgment
SRTT = αSRTT + (1 – α)RTT
RTO = β SRTT
It can’t keep up with wide fluctuations in the RTT
causing unnecessary retransmissions
RTO measurement in RFC1122
Err = RTT – SRTT
SRTT = SRTT + gErr
D = D + h( |Err| - D)
RTO = SRTT + 4D
g = gain (g)
h = 0.25 the gain for deviation

4. RTT Measurement Example
The timer increment a counter every time the 500ms
The starting Seq number is remembered

5. Karn’s Algorithm When timeout occurs and packet is retransmitted
송신 호스트는 RTT연산에 대해 혼선
Is the ACK for the first transmission or second ?
Do not update the RTT until an ACK is received for a segment that was not retransmitted
if TCP timestamp option is used, we don’t need to apply Karn’s Algorithm
Segment1,
Seq =1010
Segment1,
Seq =1010
Segment1
Retransmission
ACK, Seq=1010

6. Congestion Avoidance Algorithm
assumption of the algorithm
packet loss caused by damage is very small
the loss of a packet signals congestion
Congestion Avoidance and Slow Start algorithm
Initialization of the connection sets cwnd to one and ssthresh to 65535bytes
When congestion occurs, one-half of the current window size is saved in ssthresh. If congestion is indicated by timeout, cwnd is set to one.
When new data is acknowledged by the other end, cwnd is increased

7. Congestion Avoidance Algorithm
If cwnd is less than or equal to ssthresh, doing slow start. otherwise doing congestion avoidance.
slow start has cwnd start at one segment, and be incremented by one segment every time an ACK is received
congestion avoidance dictates that cwnd be incremented by 1/cwnd each time an ACK is received

8. Fast Retransmit and Fast Recovery
Fast Retransmission
If Sender receive three or more duplicate ACK in a row, Sender retransmit the missing segment without waiting for a retransmission timer to expire
Fast Recovery
After fast retransmission, sender do congestion avoidance
Set ssthresh to one-half of the minimum of current cwnd and receiver’s advertised window
Set cwnd to ssthresh + 3(segment size)
Each time another duplicate ACK arrives, increment cwnd by the segment size and transmit packet
When the next ACK arrives that acknowledges new data set cwnd to ssthresh

9. Congestion example segment 45 got lost Receiver send duplicated ACK
Sender get three duplicated ACKs in a row
Sender starts fast retransmission

10. Chap 1 Foundation
TCP Persist Timer

11. Chap 1 Foundation
TCP Persist Timer
When window size goes to ‘0’, Sender stops transmitting data until the window becomes nonzero
If an ACK is lost, both sender and receiver are waiting for the other
To prevent deadlock, sender use persist timer
7169:8193(1024) ack1 win 4096
8193:9217(1024) ack1 win 4096
ACK 9217 win 0
ACK 9217 win 4096
start persist timer at segment 3

12. TCP Persist Timer After segment13, sender set persist timer
If sender doesn’t receive window update when timer expire, it send window probes
bsdi
svr4
(11)
(12)
(13)
persist timer start
send window probes
(14)
(15)

13. Silly Window Syndrome It can be caused by either side
Receiver advertise small size windows
Sender transmit small amount of data
Ways to prevent silly window syndrome
Receiver side
must not advertise small window
use delay ACK
Sender side
must not transmit data until full-size or at least one-half of maximum window size
sender can send everything when we are not excepting an ACK

14. SWS example  sun.1069 bsdi.7777 PSH 1:1025(1024) ack 1, win 40963 4 5
4097:4098(1) ack 1, win 4096 6
ack 4098, win 0 7
4098:4099(1) ack 1, win 4096 8
ack 4099, win 0 9
4099:4100(1) ack 1, win 4096 10
ack 4100, win 1533 11
0.0
(0.0020)
(0.0017)
(0.0016)
(0.1649)
(4.9815)
(0.0185)
(4.9813)
(0.0187)
(4.9812)
(0.0188) 

15. SWS example 12 13 14 15 16 17 18 19 20 21
4100:5124(1024) ack 1, win 4096
ack 5124, win 509
5124:5633(509) ack 1, win 4096
ack 5633, win 0
5633:5634(1) ack 1, win 4096
ack 5634, win 1279
FIN, PSH 5634:6145(511) ack 1, win 4096
ack 6146, win 767
ack 6146, win 2816
FIN 1:1(0) ack 6146, win 4096
ack 2, win 4096 22
sun.1069
bsdi.7777
(0.0017)
(0.1983)
(4.7815)
(0.0185)
(4.9809)
(0.0191)
(0.0015)
(0.0026)
( )
( )
(0.0009) 

16. SWS example

17. SWS example

18. Chap 1 Foundation
TCP Keepalive Timer

19. TCP Keepalive Timer Introduction
If there are no data exchange between TCP connection, the connection remains established
Even if intermediate routers are crashed or lines are down, the connection remains established
Keepalive option provides capability that server can know whether client is down
If intermediate router has crashed, keepalive will think that the client is down and terminate connection

20. State of Client’s host and Server
Description
States of client and server
If there is no activity for two hours, the server sends a probe segment to the client
State of Client’s host and Server
Server’s Action
The client’s host is running and reachable from the server
Server’s TCP will reset keepalive timer for 2 hours.
If data across the TCP connection before expire, timer is reset for 2 hours again.
The client’s host has crashed and is either down or in the process of rebooting
Server will not receive response in 75 second. Server sends total 10 probes, 75 second apart. If no response, server will terminate the connection
The client’s host has just rebooted
Because client has just rebooted, server will receive response but it will be a reset. So, server terminates the connection
The client’s host is running, but unreachable from the server
Same as client’s down. Server terminates the connection after 10 probes, 75 second apart.

21. - disconnect cable –(4 hours later) second keepalive probe
Keepalive Example
Other End Crashes
bsdi % sock –K svr4 echo
hello, world
- disconnect cable –(4 hours later)
read error: Connection timed out
first keepalive probe
(2 hours later)
second keepalive probe
(4 hours later)

22. read error: Connection reset by peer
Keepalive Example
Other End Crashes and Reboot
bsdi % sock –K svr4 echo
hi there
read error: Connection reset by peer
first keepalive probe
(2 hours later)

23. - link is down (3 hours later) - second keepalive probe
Keepalive Example
Other End is Unreachable
slip % sock –K vangogh.cs.berkeley.edu echo
testing
- link is down (3 hours later) -
read error: No route to host
first keepalive probe
(2 hours later)
second keepalive probe
(4 hours later)
Client
Server
Terminate the connection
Router
(alive)
(down)