Transmission Control Protocol
(TCP)
Chapter 15
TCP/IP Protocol Suite
2
TCP Services
Process-to-process communication
Stream delivery service
Connection-oriented service
Reliable service
Error control
Flow control
Congestion control
3
Stream Delivery Service
Buffering handles disparity between speed of
sending and receiving processes
Segments created from the bytes in buffers
Segment N Segment 1
H H
Stream of bytes
4
TCP Segment Format
5
Control Field
6
TCP Checksum
Checksum in TCP is mandatory
7
TCP Connection Establishment
Three-way handshaking
Passive
Active open
open seq: 8000
UAPRS F
SYN
seq: 15000
Connection ack: 8001
opened nd: 5000
U A P R S F rw
SYN + ACK
seq: 8000
ack: 15001
UAPRS F
rwnd: 10000
ACK
See RFC 793
8
TCP Data Transfer
seq: 8001
Send
request ack: 15001
UAPRS F
Data Receive
bytes: 8001-9
000
Send seq: 9001
request ack: 15001
UAPRS F
Data Receive
bytes: 9001-1
0000
seq: 15001 Send
ack: 10001 request
UAPRSF
Data
17000
bytes: 15001-
seq: 10000
ack: 17001
UAPRS F
rwnd:10000
Connection Termination
9
Pushing Data and Urgent Data
Pushing
Send immediately, not to wait for window to fill
Set push bit PSH
Deliver as soon as possible, not to wait for data
Urgent
Insert urgent data at beginning of segment
Set urgent bit URG
Urgent pointer defines end of urgent data
Inform receiving application to take action
10
TCP Connection Termination
Three-way handshaking
See RFC 793
11
Half-Close
12
TCP Connection Reset: Denying a Connection
13
TCP Connection Reset: Aborting a Connection
14
Windows in TCP
TCP uses two windows for each direction of
data transfer
Send window
Receive window
15
Send Window
Closes when acknowledgement received
Opens/shrinks when rwnd advertised by receiver 16
Receive Window
Closes when more bytes arrive from sender
Opens when more bytes pulled by process 17
Flow Control
To avoid overwhelming receiver with data
Implemented as sliding window
Window size determined by receiver (rwnd)
Messages Flow control
are pushed
1 5 3 Messages
feedback are pulled
2
Segements are pushed
4
Flow control feedback
18
Example of Flow Control
19
Shrinking of Windows
Receive window cannot shrink
Send window can shrink
To prevent shrinking of send window
20
Example of Flow Control
Last
New
21
Example of Flow Control
Last
New
22
Example of Flow Control
Last
New
23
Shrinking of Windows – Example
Send
Sender sent bytes 206 to 214
Sender received Ack=210, rwnd=4
210
Send
24
Window Shutdown
Receiver can shut down window (rwnd=0)
Sender stops sending data until new rwnd arrived
Sender always sends 1-byte segment (probing) to
prevent deadlock
25
Silly Window Syndrome
Sending very small segments caused by
Sender creates data slowly
Receiver consumes data slowly
Inefficient use of network capacity
26
Solution to Syndrome Created by Sender
Nagle’s Algorithm:
1. Send first segment even if it is only 1 byte
2. Accumulate data in output buffer and wait
until receiving acknowledgement or enough
data filled maximum-size segment
3. Repeat step 2 for rest of transmission
27
Solution to Syndrome Created by Receiver
Clark’s Solution
Send acknowledgement but announce
rwnd=0 until enough space for maximum-
size segment or half of receive buffer empty
Delayed Acknowledgement
Delay sending acknowledgement until decent
space available in incoming buffer
Not delayed more than 500 ms
28
Congestion Control
To avoid and detect congestion in network
Sender’s window size determined by
Receiver (rwnd)
Congestion in network (cwnd)
Actual window size = minimum (rwnd, cwnd)
Based on three phases
Slow start
Congestion avoidance
Congestion detection
29
Slow Start: Exponential Increase
Sender starts with cwnd = 1 MSS (maximum segment size)
cwnd size increases by 1 MSS for each acknowledged segment
cwnd size increases exponentially until it reaches threshold
1
cwnd
RTT
2
cwnd
RTT
4
cwnd
RTT
8
cwnd 30
Congestion Avoidance: Additive Increase
When cwnd size reaches threshold, congestion avoidance begins
cwnd size increases by 1 MSS when whole window acknowledged
cwnd size increases additively until congestion detected
31
Congestion Detection: Multiplicative Decrease
Congestion detected by retransmission
If congestion detected by time-out
Set threshold to half current window size
Set cwnd size to 1 MSS
Start slow start phase
If congestion detected by 3 duplicate ACKs
Set threshold to half current window size
Set cwnd size to threshold
Start congestion avoidance phase
32
Congestion Example
Threshold = 1/2 window
cwnd = 1
Threshold = 1/2 window
cwnd = Threshold
33
TCP Timers
34
Retransmission Timer
Value of retransmission time-out (RTO) based on:
Smoothed RTT (round trip time)
RTT Deviation
RTTD = (1 - β) RTTD + β . |RTTS – RTTM| (β = 1/4)
RTTS = (1 - α) RTTS + α . RTTM (α = 1/8)
RTO = RTTS + 4 . RTTD
Karn’s Algorithm
Do not consider RTT of retransmitted segment
Do not update RTT until you send segment and receive ACK
TCP uses exponential backoff strategy:
RTOnew = 2 x RTO
35
Retransmission Timer – Example
0.8125 4.875
See RFC 6298
36
Karn’s Algorithm – Example
0.8125 4.875
4.875 = 9.75
9.75
1.203 6.73
37
Persistence Timer
To deal with zero-window-size advertisement
(deadlock)
Set to value of retransmission time
When time-out, TCP sends 1-byte probe segment
If response not received, another probe sent and
timer doubled until 60s
38
Keepalive Timer
To prevent long idle connection
After time-out (2 hours)
Server sends probe segment to client
If no response after 10 probes (every 75s),
server terminates connection
39
TIME-WAIT Timer
Used during connection termination
If last ACK segment lost
If duplicate segment from one connection appears in next one
Twice maximum segment lifetime (MSL=30s - 60s)
40
TCP Options
41
End-of-Option (EOP) Option
42
No-Operation (NOP) Option
43
Maximum-Segment-Size (MSS) Option
Each end defines MSS for segments it will receive
during connection (default = 536 bytes)
Determined during connection establishment
44
Window-Scale-Factor Option
To increase window size defined in TCP header
Determined during connection establishment
new window size =
window size defined in header x 2window scale factor
45
Timestamp Option
To measure round-trip time (RTT)
Protection against wrapped sequence numbers (PAWS)
Identity of segment defined as combination of
timestamp and sequence number
46
Timestamp Option – Measuring RTT
47
SACK-Permitted and SACK Options
Allow receiver to send selective ACK (SACK)
SACK-permitted exchanged during connection
establishment to use SACK during data transfer
SACK includes list of
Out-of-order segments
Duplicate segments
48
SACK-Permitted and SACK Options
49
SACK Option – Reporting Out-of-Order
50
SACK Option – Reporting Out-of-Order and Duplicate
51
SACK Option – Reporting Duplicate of Out-of-Order
52
