1. The Transmission Control Protocol (TCP)
Application Services
(Telnet, FTP, , WWW)
Reliable Stream
Transport (TCP)
Connectionless Packet Delivery Service
(IP)
Unreliable Transport
Service (UDP)

2. The Transmission Control Protocol (TCP)
TCP is a protocol that specifies:
How to distinguish among multiple destinations on a given machine
How to initiate and terminate a stream transfer
Format of the data and acknowledgments that two computers exchange to achieve a reliable transfer
Procedures the computers use to ensure that the data arrives correctly

3. ACKNOWLEDGMENT NUMBER
TCP Segment Format
TCP divides the data stream into segments for transmission:
...
CHECKSUM URGENT POINTER
OPTIONS (IF ANY) PADDING
DATA
HLEN RESERVED CODE BITS WINDOW
ACKNOWLEDGMENT NUMBER
SOURCE PORT DESTINATION PORT
SEQUENCE NUMBER

4. TCP Segment Header Fields
Code bits - identify the contents of the segment:
Window - how much data the sender is willing to accept (flow control)
Urgent pointer - specifies the position in the segment where urgent data ends

5. Out of Band Data TCP provides a mechanism to handle urgent data
Urgent data is received before octets already in the stream
Sender:
Sets urgent bit in segment header
Puts urgent data at the beginning of the data field
Sets urgent pointer to the end of the urgent data
Receiver:
Notified of the urgent data as soon as it arrives
Enters “urgent mode” until all urgent data has been consumed
Returns to “normal mode”

6. TCP Segment Header Fields (cont)
Checksum - used to verify the integrity of the segment
Computed over:
Segment header
Segment data
Pseudo header
1’s complement addition algorithm
Options - signal various options
Maximum segment size

7. TCP - Acknowledgments and Retransmission
Uses cumulative acknowledgment scheme:
ACK 7 = “all octets up to but not including number 7 have been received correctly”
Advantages
ACKs are easy to generate and unambiguous
Lost acknowledgments do not force retransmission
Disadvantages
Sender does not receive information about all successful transmissions

8. TCP - Timeout and Retransmission
For each segment sent:
Start timer and wait for acknowledgment
Retransmit if timer expires
TCP uses an adaptive retransmission algorithm because internet delays are so variable
Round trip time of each connection is recomputed every time an acknowledgment arrives
Timeout value is adjusted accordingly

9. TCP - Timeout and Retransmission (cont)
How should the most recent round trip sample (RTS) effect the round trip time (RTT)?
New RTT = (a * RTT) + ((1-a) * RTS)
a = 0
a = 1
How should RTT be used to compute timeout?
Timeout = b * RTT
b = 1
b > 1

10. Accurate Measurement of Round Trip Samples
Simple answer:
Difference in time segment sent and time ACK received
But what about retransmissions?
Does ACK correspond to first or second copy of segment sent?
Assuming ACK is for the earliest transmission causes problems
Assuming ACK is for the latest transmission causes problems
TCP acknowledgments are ambiguous

11. Accurate Measurement of Round Trip Samples
To avoid problems with ambiguous ACKs:
TCP should not update the RTT for retransmitted segments
Problem: if timeout value is too small all segments will cause retransmissions and RTT will never be updated
Timer back-off: if the timer expires and causes a retransmission TCP increases the timeout
New timeout = c * timeout

12. Karn’s Algorithm
“When computing the round trip estimate, ignore samples that correspond to retransmitted segments, but use a backoff strategy, and retain the timeout value from a retransmitted packet for subsequent packets until a valid sample is obtained.”

13. Responding to High Variance in Delay
Often a good estimate of the round trip time is not very useful because internet delays tend to have a high variance
Most TCP implementations estimate:
Average round trip time
Variance
Using the variance
Timeout = b * RTT

14. Responding to Congestion
Endpoints cannot know the details of where in the internet congestion has occurred or why
Congestion will usually lengthen delays
TCP’s response to lengthened delays (retransmission) can cause congestion collapse
Instead, TCP must reduce transmission rates when congestion occurs

15. Responding to Congestion (cont)
Augment the TCP transmission window with a congestion window:
Window=min(receiver advertisement,congestion window)
Multiplicative Decrease Congestion Avoidance:
Upon loss of a segment reduce the congestion window by half (down to a minimum of one segment)
Backoff the retransmission timer exponentially

16. Responding to Congestion (cont)
Slow-start (additive) recovery:
When starting traffic on a new connection or increasing traffic after a period of congestion
Start the congestion window at the size of a single segment and increase the congestion window by one segment each time an ACK arrives
Congestion avoidance phase (during recovery):
Once the congestion window reaches half of its original size before congestion occurred
Increase the size of the congestion window by 1 only if all segments in the window have been acknowledged

17. TCP Timeout and Retransmission - Summary
What do we get for all this stuff:
Slow-start increase
Multiplicitive decrease
Congestion avoidance
Measurement of RTT and variance
Exponential timer backoff
Dramatic improvement of TCP performance without adding significant computational overhead

18. Establishing a TCP Connection
The 3-way handshake
Guarantee that both sides are ready for connection
Allows both sides to agree on initial sequence numbers
Site 1
Network
Site 2
Send SYN seq=x
Send SYN seq=y, ACK x+1
Send ACK y+1
Receive SYN&ACK
Receive ACK
Receive SYN

19. Closing a TCP Connection
Applications should close a connection when they have no more data to transmit
Connection can be closed in either one or both directions
Site 1 finishes transmitting data and waits for ACK from site 2
Site 1 transmits a segment with the FIN bit set
Site 2 acknowledges the FIN segment
Site 2 notifies the application that no more data is coming
Data can still be transmitted from site 2 to site 1
Site 1 will still receive and acknowledge data from site 2
Eventually, site 2 will finish transmitting and close its connection
Both endpoints delete record of the connection

20. Closing a TCP Connection (cont)
Site 1
Network
Site 2
(app closes connection)
Send FIN seq=x
Receive FIN Send ACK x+1
(inform application)
Receive ACK
(app closes connection)
Send FIN seq=y, ACK x+1
Receive FIN&ACK
Send ACK y+1
Receive ACK

21. TCP Connection Reset Applications normally close connections
Sometimes abnormal conditions arise that break a connection
Broken connections can be reset:
Site 1 sends a segment with the RST bit set
Site 2 receives segment and aborts the connection
Transfers in both directions cease immediately
Resources for the connection are released
Applications programs are informed

22. Forcing Data Delivery
TCP divides the stream of octets into segments for transmission
This improves efficiency since octets can be buffered until a good-sized segment can be sent
TCP provides a push operation for applications that want to force delivery of octets
Set PSH bit
Send segment

23. Reserved TCP Port Numbers
Like UDP:
Static port bindings for commonly used services
Ports are reserved
Dynamic port bindings
Port numbers over 1024
Port numbers for services accessible by both UDP and TCP usually match
ECHO (7)
TIME (37)

24. Reserved TCP Port Numbers
Port number: Service:
0 Reserved
7 Echo
17 Quote of the day
21 FTP
23 TELNET
25 SMTP
37 Time
79 Finger
80 HTTP
119 NNTP

25. TCP Performance Silly Window Syndrome Sender generates data quickly
Receiver reads incoming data one octet at a time
Sender
Receiver

26. TCP Performance (cont)
Silly Window Syndrome
Each ACK advertises a small amount of space
Each segment carries a small amount of data
Problems:
Poor use of network bandwidth
Unnecessary computational overhead

27. TCP Performance (cont)
Avoiding Silly Window Syndrome
Use heuristics at sender to avoid transmitting a small amount of data in each segment
Use heuristics at receiver to avoid sending small window advisements
Receive-side silly window avoidance
Monitor receive window size
Delay advertising an increase until a “significant” increase is possible
“Significant” = min(half the window, maximum segment size)

28. Receive-Side Silly Window Avoidance Example
Receive 6 octets, send ACK 7 with window advisement of 0
Application reads one octet
Application reads one octet
Application reads one octet
Send window advisement of 3, receive 3 octets

29. Receive-Side Silly Window Avoidance
Two approaches:
Receiver can ACK received octets but does not advertise an increase in its window until the increase is significant
Receiver can not send ACKs when the window is not large enough to advertise
Advantages/disadvantages?

30. Send-Side Silly Window Avoidance
Goal: avoid sending small segments
Application can generate data in small blocks
TCP must collect data sent by application into a single large segment (clump) for transmission
TCP must delay sending a segment until it contains a reasonable amount of data
How long should TCP wait before transmitting data?

31. Send-Side Silly Window Avoidance (cont)
The Nagle Algorithm:
Application generates data to be sent over a connection that has already transmitted some data
If all previous transmissions have been acknowledged send the data immediately
If any ACKs are still pending do not transmit until:
Maximum segment size is reached, or
An ACK arrives
Self-clocking - does not compute delays
Applies even if the application requests a push

32. TCP Summary Provides reliable stream delivery service
Full duplex
Out-of-band for urgent data
Makes efficient use of the network
Piggybacking
Sliding windows
Efficiency
End-to-end flow control
Acknowledgment and retransmission
Congestion recovery/avoidance