TCP and UDP

1
TCP and UDP
2
The Internet Transport Layer

• The Internet supports two transport layer
  protocols
• The Transport Control Protocol (TCP) for reliable
  service
• The Unreliable Datagram Protocol (UDP)

3
TCP

• TCP provides the end-to-end reliable connection
  that IP alone cannot support
• The TCP connection primitive is known as the
  socket
• A socket may be viewed as a bi-directional pipe
  between two hosts
• Connecting a socket requires a destination IP
  address and a port number

4
TCP Ports

• Historical convention was that ports numbered
  less than 1024 are reserved for standard services
  like
• Telnet (port 23)
• FTP (21 control, 20 data)
• Email (port 25)
• Secure Shell (ssh, port 22),
• etc. (See a UNIX /etc/services file for more)

5
The TCP Protocol

• Frame format
• Connection management
• Flow control
• Congestion control

6
TCP Frame Format
32 bits
Source Port
Destination Port
Sequence Number
Acknowledgement number
Window Size
HL
F I N
S Y N
R S T
P S H
A C K
U R G
Checksum
Urgent Pointer
Options (0 or more 32-bit words)
Data
7
TCP Frame Fields

• Source Destination Ports
• 16 bit port identifiers for each packet
• Sequence number
• The packets unique sequence ID
• Acknowledgement number
• The sequence number of the next packet expected
  by the receiver

8
TCP Frame Fields (contd)

• Window size
• Specifies how many bytes may be sent after the
  first acknowledged byte
• Checksum
• Checksums the TCP header and IP address fields
• Urgent Pointer
• Points to urgent data in the TCP data field

9
TCP Frame Fields (contd)

• Header bits
• URG Urgent pointer field in use
• ACK Indicates whether frame contains
  acknowledgement
• PSH Data has been pushed. It should be
  delivered to higher layers right away.
• RST Indicates that the connection should be
  reset
• SYN Used to establish connections
• FIN Used to release a connection

10
TCP Connection Establishment

• Three-way Handshake

Host A
Host B
SYN (seqx)
SYN (seqy, ACKx1)
SYN (seqx1, ACKy1)
11
TCP Connection Tear-down

• Two double handshakes

Host A
Host B
FIN (seqx)
ACK (ACKx1)
A-gtB torn down
FIN (seqy)
ACK (ACKy1)
B-gtA torn down
12
TCP Flow Control

• TCP uses a modified version of the sliding window
• In acknowledgements, TCP uses the Window size
  field to tell the sender how many bytes it may
  transmit
• TCP uses bytes, not packets, as sequence numbers

13
TCP Flow Control (contd)
Important information in TCP/IP packet headers
Number of bytes in packet (N)
Sequence number of first data byte in packet (SEQ)
N
SEQ
Send
Window size at the receiver (WIN)
Sequence number of next expected byte (ACK)
ACK bit set
ACK
WIN
Recv
Contained in IP header
Contained in TCP header
14
Hypothetical TCP session

• (1)remus tcpdump -S host scullyKernel
  filter, protocol ALL, datagram packet
  sockettcpdump listening on all devices
• 151522.152339 eth0 gt remus.4706 gt
  scully.echo S 12642965041264296504(0) win 32120
  ltmss 1460,sack OK,timestamp 71253512 0,nop,wscale
  0gt 151522.153865 eth0 lt scully.echo gt
  remus.4706 S 875676030875676030(0) ack
  1264296505 win 8760 ltmss 1460gt151522.153912
  eth0 gt remus.4706 gt scully.echo .
  12642965051264296505(0) ack 875676031 win 32120

remus telnet scully 7 A ltreturngt A
15
Example TCP session
Timestamp
Source IP/port
Dest IP/port
Packet 1 151522.152339 eth0 gt remus.4706 gt
scully.echo S 12642965041264296504(0) win 32120
ltmss 1460,sackOK,timestamp 71253512 0,nop,wscale
0gt (DF)
Flags
Packet 2 151522.153865 eth0 lt scully.echo gt
remus.4706 S 875676030875676030(0) ack
1264296505 win 8760 ltmss 1460)
Options
Packet 3 151522.153912 eth0 gt remus.4706 gt
scully.echo . 12642965051264296505(0) ack
875676031 win 32120
Window
Start Sequence Number
Acknowledgement Number
End Sequence Number
16
TCP data transfer
Packet 4151528.591716 eth0 gt remus.4706 gt
scully.echo P 12642965051264296508(3) ack
875676031 win 32120
data
Packet 5 151528.593255 eth0 lt scully.echo gt
remus.4706 P 875676031875676034(3) ack
1264296508 win 8760
bytes
17
TCP Flow Control (contd)
Receivers buffer
Receiver
Sender
Application does a 2K write
0
4K
Empty
ACK 2048 WIN 2048
Application does a 3K write
Full
Sender is blocked
Application reads 2K
ACK 4096 WIN 0
ACK 4096 WIN 2048
Sender may send up to 2K
18
TCP Flow Control (contd)
Piggybacking Allows more efficient
bidirectional communication
Data from A to B
ACK for data from B to A
N
SEQ
ACK
WIN
A
B
N
SEQ
ACK
WIN
Data from B to A
ACK for data from A to B
19
TCP Retransmission

• When a packet remains unacknowledged for a period
  of time, TCP assumes it is lost and retransmits
  it
• TCP tries to calculate the round trip time (RTT)
  for a packet and its acknowledgement
• From the RTT, TCP can guess how long it should
  wait before timing out

20
Round Trip Time (RTT)
Time for data to arrive
Network
Time for ACK to return
RTT Time for packet to arrive at destination
Time for ACK to return from
destination
21
RTT Calculation
Receiver
Sender
0.9 sec
RTT
ACK 2048
2.2 sec
RTT 2.2 sec - 0.9 sec. 1.3 sec
22
Smoothing the RTT measurement

• First, we must smooth the round trip time due to
  variations in delay within the network
• SRTT a SRTT (1-a) RTTarriving ACK
• The smoothed round trip time (SRTT) weights
  previously received RTTs by the a parameter
• a is typically equal to 0.875

23
Calculating the Retransmission Timeout Interval

• The timeout value is then calculated by
  multiplying the smoothed RTT by some factor
  (greater than 1) called b
• Timeout b SRTT
• This coefficient of b is included to allow for
  some variation in the round trip times.

24
Smoothing the RTT measurementExample
Initial SRTT 1.50 a 0.875, b 4.0
RTT Meas.
SRTT
Timeout
1.5 s
1.50
b1.50 6.00
1.0 s
1.50a 1.0(1- a) 1.44
b1.44 5.76
2.2 s
1.44a 2.2(1- a) 1.54
b1.54 6.16
1.0 s
1.54a 1.0(1- a) 1.47
b1.47 5.88
0.8 s
1.47a 0.8(1- a) 1.39
b1.39 5.56
3.1 s
2.0 s
25
Problem with RTT Calculation
Receiver
Sender
Sender Timeout
RTT?
ACK 2048
RTT?
26
Karns Algorithm

• Retransmission ambiguity
• Measure RTT from original data segment
• Measure RTT from most recent segment
• Either way there is a problem in RTT estimate
• One solution
• Never update RTT measurements based on
  acknowledgements from retransmitted packets
• Problem Sudden change in RTT can cause system
  never to update RTT
• Primary path failure leads to a slower secondary
  path

27
Karns algorithm

• Use back-off as part of RTT computation
• Whenever packet loss, RTO is increased by a
  factor
• Use this increased RTO as RTT estimate for the
  next segment (not from srtt)
• Only after an acknowledgment received for a
  successful transmission is the timer set to new
  RTT obtained from srtt

28
Another Problem with RTT Calculation

• RTT measurements can sometimes fluctuate severely
• smoothed RTT (SRTT) is not a good reflection of
  round-trip time in these cases
• Solution Use Jacobson/Karels algorithm
• Error RTT - SRTT
• SRTT SRTT (a Error)
• Dev Dev h(Error - Dev)
• Timeout SRTT (b Dev)

29
Jacobson/Karels AlgorithmExample
Initial SRTT 1.50, Dev 0 a 0.125, d 0.25,
b 4.0
Error RTT - SRTT SRTT SRTT (a Error) Dev
Dev d (Error - Dev) Timeout SRTT (b
Dev)
RTT Meas.
SRTT
Error
Dev.
Timeout
1.5 s
1.50
0.0
0.00
1.50
1.0 s
1.44
-0.50
0.13
1.94
2.2 s
1.54
0.76
0.28
2.67
1.0 s
1.47
-0.54
0.35
2.85
0.8 s
1.39
-0.67
0.43
3.09
3.1 s
2.0 s
30
TCP Congestion Control

• Recall Network layer is responsible for
  congestion control
• However, TCP/IP blurs the distinction
• In TCP/IP
• the network layer (IP) simply handles routing and
  packet forwarding
• congestion control is done end-to-end by TCP

31
Self-Clocking Model
Bottleneck link
Pr
Fast link
Pb
1. Send Burst
2. Receive data packet
5. Send a data packet
Data
Receiver
Sender
Acks
4. Receive Acknowledgement
3. Send Acknowledgement
Ab
Ar
Ar
Given Pb Pr Ar Ab Ar Sending a packet on
each ACK keeps the bottleneck link busy
32
TCP Congestion Control

• Goal achieve self-clocking state
• Even if dont know bandwidth of bottleneck
• Bottleneck may change over time
• Slow-start ramps up to the bottleneck limit
• Packet loss signals we passed bandwidth of
  bottleneck
• Congestion Avoidance tries to maintain self
  clocking mode once established

33
TCP Congestion Window

• TCP introduces a second window, called the
  congestion window
• This window maintains TCPs best estimate of
  amount of outstanding data to allow in the
  network to achieve self-clocking
• To determine how many bytes it may send, the
  sender takes the minimum of the receiver window
  and the congestion window
• Example
• If the receiver window says the sender can
  transmit 8K, but the congestion window is only
  4K, then the sender may only transmit 4K
• If the congestion window is 8K but the receiver
  window says the sender can transmit 4K, then the
  sender may only transmit 4K

34
TCP Congestion Control

• The TCP Congestion Control algorithm makes use
  of
• Slow Start
• Congestion Avoidance (Linear Increase Thresholds)

35
TCP Slow Start

• TCP defines the maximum segment size as the
  maximum size a TCP packet can be (including
  header)
• TCP Slow Start
• Congestion window starts small, at 1 segment size
• Each time a transmitted segment is acknowledged,
  the congestion window is increased by one maximum
  segment size
• On each ack, cwndcwnd 1

36
TCP Slow Start (contd)
Congestion Window Size
Event

1K A sends 1 segment to B B ACKs the
segment 2K A sends 2 segments to B B ACKs both
segments 4K A sends 4 segments to B B ACKs all
four segments 8K A sends 8 segments to B B
ACKs all eight segments 16K and so on
37
TCP Slow Start (contd)

• Congestion window size grows exponentially (i.e.
  it keeps on doubling)
• Packet losses indicate congestion
• Packet losses are determined by using timers at
  the sender
• When a timeout occurs, the congestion window is
  reduced to one maximum segment size and
  everything starts over

38
TCP Slow Start (contd)
Timed out Transmissions
Congestion window
Transmission Number
1 Maximum Segment Size
39
TCP Slow Start (contd)

• TCP Slow Start by itself is inefficient
• Although the congestion window builds
  exponentially, it drops to 1 segment size every
  time a packet times out
• This leads to low throughput

40
TCP Linear Increase Threshold

• Establish a threshold at which the rate increase
  is linear instead of exponential to improve
  efficiency
• Algorithm
• Start the threshold at 64K (ssthresh)
• Start the congestion window size at 1 segment
  size
• Increase the congestion window size exponentially
  using slow start until the threshold is reached
• Once the threshold is passed, only increase the
  congestion window size by 1 segment size for each
  congestion window of data transmitted
• For each ack received, increment cwnd as follows
• cwnd cwnd (mssmss)/cwnd
• Note that cwnd is a multiple of mss for some k
  (say kmss)
• During the linear phase, if we get k acks then
  cwnd increases by 1 mss
• If a timeout occurs, reset the congestion window
  size to 1 segment and set threshold to 1/2 of
  MIN(sliding window, congestion window)

41
TCP Linear Increase Threshold
Example Maximum segment size 1K Assume
SSthresh32K
Timeout occurs when MIN(sliding window,
congestion window) 40K
Congestion window
Thresholds
40K
32K
20K
1K
Transmission Number
42
TCP Fast Retransmit

• Another enhancement to TCP congestion control
• Idea When sender sees 3 duplicate ACKs, it
  assumes something went wrong
• The packet is immediately retransmitted instead
  of waiting for it to timeout
• Why?
• Note that acks sent by the receiver when it
  receives a packet
• Dup ack implies something is getting through
• Better than time out

43
TCP Fast RetransmitExample
Receiver
Sender
MSS 1K
ACK of new data
ACK 2048 WIN 31K
Duplicate ACK 1
ACK 2048 WIN 30K
Duplicate ACK 2
ACK 2048 WIN 29K
Fast Retransmit occurs (2nd packet is
now retransmitted w/o waiting for it to timeout)
Duplicate ACK 3
ACK 2048 WIN 28K
ACK 2048 WIN 27K
ACK 7168 WIN 26K
44
TCP Fast Recovery

• Yet another enhancement to TCP congestion control
• Idea Dont do a slow start after a fast
  retransmit
• Instead, use this algorithm
• Drop threshold to 1/2 of MIN(sliding window,
  congestion window)
• Set congestion window to threshold 3 MSS
• (ssthresh 3mss)
• For each duplicate ACK (after the fast
  retransmit), increment congestion window by MSS
• When next non-duplicate ACK arrives, set
  congestion window equal to the threshold

45
TCP Fast RecoveryExample
Sender
SW29K,TH15K, CW20K
Continuing with the Fast Retransmit Example...
SW28K,TH15K, CW20K
ACK 2048 WIN 28K
Fast Retransmit Occurs
MSS1K Sliding Window (SW) Congestion Threshold
(TH) Congestion Window (CW)
SW28K, TH10K, CW13K
ACK 2048 WIN 27K
SW27K, TH10K, CW14K
ACK 7168 WIN 26K
SW26K, TH10K, CW10K
46
Resulting TCP Sawtooth
In steady state, window oscillates around the
bottlenecks capacity (I.e. number of
outstanding bytes in transit)
Congestion window
Slow Start
Capacity
Linear Mode
40K
Sawtooth
32K
20K
1K
Transmission Number
47
TCP Recap

• Timeout Computation
• Timeout is a function of 2 values
• the weighted average of sampled RTTs
• The sampled variance of each RTT
• Congestion control
• Goal Keep the self-clocking pipe full in spite
  of changing network conditions
• 3 key Variables
• Sliding window (Receiver flow control)
• Congestion window (Sender flow control)
• Threshold (Senders slow start vs. linear mode
  line)
• Slow start
• Add 1 segment for each ACK to the congestion
  window
• -Doubles the congestion windows volume each
  RTT

48
TCP Recap (cont)

• Linear mode (congestion avoidance)
• Add 1 segments worth of data to each
  1/congestion window
• Adds 1 segment per RTT

49
UDP

• UDP is an unreliable transport protocol that can
  be used in the Internet
• UDP does not provide
• flow or error control
• connection management
• guaranteed in-order packet delivery
• UDP is almost a null transport layer

50
UDP Frame Format
32 bits
Source Port
Destination Port
UDP length
UDP checksum (optional)
Data
51
Why UDP?

• No connection needs to be set up
• Throughput may be higher because UDP packets are
  easier to process, especially at the source
• The user doesnt care if the data is transmitted
  reliably
• The user wants to implement his or her own
  transport protocol