Section 11: Transport Layer Protocols and Interfaces

1
Section 11Transport Layer Protocols and
Interfaces

• In this section
• The Transmission Control Protocol (TCP)
• Service model
• Message format
• Connection and Disconnection
• State machine
• Flow control
• Timer management
• The User Datagram Protocol (UDP)
• Sockets

2
TCP - Transmission Control Protocol

• Provides reliable end-to-end byte stream over
  unreliable network.
• Designed to dynamically adapt to properties of
  network, and for robust handling of failures.
• Initial definition in standard RFC 793.
• TCP entity accepts user data from local
  processes, splits into pieces no larger than 64K
  octets (usually about 1500 octets in practice),
  and sends each piece as a separate IP datagram.
• When IP datagrams containing TCP information
  arrive, TCP entity receives them, and
  reconstructs the original byte stream.
• TCP times out, retransmits IP datagrams as
  needed, and reorders on reception.

3
The TCP Service Model

• Sender and receiver use sockets as end points.
• Sockets have an address consisting of IP address
  for the machine (such as 127.1.1.1), plus a 16
  bit number local to the host called a port.
• Connection is established between sockets.
• Port numbers 0 through 1023 are well known, and
  are reserved for standard services which are
  expected to be found at one of these ports.
• Examples FTP daemons at ports 20 and 21, HTTP
  server at port 80
• All TCP connections are full duplex, point to
  point.

4
The TCP Service Model - 2

• Connection is a stream of octets, not messages.
• Four messages of 512 octets may be sent by TCP as
  any combination of pieces totaling 2048 octets
• 1 2048 octets
• 2 1024 octets
• 4 512 octets
• 8 256 octets
• ...

5
The TCP Service Model - 3

• When application sends data to TCP, it may not be
  sent immediately
• TCP may decide to collect more data before
  transmission.
• TCP push flag is used to request no transmission
  delay.
• TCP urgent flag immediately sends all pending
  data for connection.
• Receiver immediately sends an interrupt to its
  user.
• On reception, data is buffered until retrieved by
  a request from the socket.

6
The TCP Protocol

• 32 bit sequence numbers used.
• Separate fields for acknowledgments and window
  mechanism.
• TCP entities exchange data in the form of
  segments
• Fixed 20 octet header, plus optional part,
  followed by zero or more data octets.
• TCP software determines segment size.
• Restrictions on maximum segment size
• IP payload maximum 65,535 octets, including IPs
  20 octet header.
• MTU at data link layer.
• TCP uses sliding window protocol
• Sender starts timer after segment sent.
• Receiver sends back segment (which may or may not
  contain data) with acknowledgment number being
  the next sequence number it expects to receive.
• Sender retransmits on timeout.

7
TCP must handle...

• Fragmentation only part of a transmission
  arrives.
• Out-of-order segments.
• Duplicates segments delayed long enough in
  transit that sender retransmits.
• Duplicates fragmentation retransmission may
  be fragmented differently.
• Network congestion.

8
The TCP Segment Header
Source Port
Destination Port
Sequence Number
Acknowledgment Number
Window size
FIN
S Y N
Header Length
A C K
P S H
R S T
U R G
Checksum
Urgent Pointer
Options (0 or more 32 bit words)
Data (optional)
32 bits
Header length 4 bits ACK / PSH / RST / SYN /
FIN / URG 1 bit Other fields 16 or 32 bits
unless noted Unused 6 bits
9
TCP Header Fields - 1

• Source and destination ports 16 bit address of
  local port.
• Sequence and acknowledgment numbers
• Every octet is numbered in a TCP stream.
• Acknowledgment number is next octet number
  expected.
• 32 bits each.
• Header length
• Needed because options field can vary in length.
• Number of 32 bits words in header.
• URG set to 1 if urgent pointer in use
• Pointer indicates offset from current sequence
  number at which urgent data is found.
• Facility provided in lieu of interrupts.
• PSH Set to 1 to indicate pushed data.
• Receiver is requested to immediately send data to
  user, instead of storing in buffer.

10
TCP Header Fields - 2

• ACK Set to 1 to indicate acknowledgment number
  is valid
• If 0, no acknowledgment in this segment.
• SYN used to establish connections
  (synchronize).
• SYN 1, ACK 0 in connection request.
• SYN 1, ACK 1 in connection acceptance.
• SYN 0, ACK 1 to acknowledge connection
  acceptance
• FIN set to 1 to indicate end of user data
  (finished).
• Used to close connection in sending direction.
• May continue to receive data.
• RST Set to 1 to indicate reset.
• Host has become confused due to crash or for
  other reason.
• Also used to reject a connection, or refuse an
  invalid segment.

11
TCP Error Checking
TCP Pseudo-header
Source IP address
Destination IP address
TCP segment length
00000000
Protocol 6
32 bits

• Checksum field provides error detection
  information for TCP segment header, plus the
  pseudo-header shown above.
• Checksum computation
• Set checksum field to all zeros.
• Pad user data with extra 0 octet, if needed, so
  that user data has an even number of bytes.
• Add all 16 bit words in 1s complement, and take
  1s complement of the sum.
• When receiver performs this computation,
  including checksum field, result should be 0.

12
TCP Connection and Disconnection

• Initial sequence number on connection
• Based on clock that ticks every 4 ?sec.
• Maximum packet lifetime MPL seconds (often 30s
  to 60s).
• When host crashes, it must delay 2MPL seconds to
  let outstanding packets be cleared.
• Disconnection
• Segment with FIN bit set indicates end of user
  data.
• Connection is then shut down in that direction.
• Data can continue to be sent in the other
  direction.
• If no acknowledgment for a segment with FIN is
  received in 2MPL, connection is released.

13
TCP connection Three-way Handshake
Host 1
Host 2
ack
seq
SYN Propose sequence number x
SYN ACK Propose sequence number Y, and
confirm X
ACK Confirm sequence number Y
Normal, successful connection
14
TCP simultaneous disconnection
Host 1
Host 2
FIN Close connection for sending at Host 1.
FIN ACK Close connection for sending at host
2, and confirm FIN reception
ACK Confirm FIN from Host 2
15
TCP Half-close
Host 1
Host 2
FIN Close connection for sending at Host 1.
ACK Confirm FIN reception
(additional data transfer)
FIN Close connection for sending at host 2
ACK Confirm FIN reception
16
Sequencing of Control Segments

• A segment with the SYN bit on consumes one
  sequence number, although there is no data.
• A segment with the FIN bit on consumes one
  sequence number, although there is no data.
• A segment with the ACK bit on, the SYN and FIN
  bits off, and that does not carry data, does not
  consume a sequence number.

17
TCP Connection Management States
18
TCP Connection Establishment
CONNECT / SYN
Closed
LISTEN / -
CLOSE / -
Listen
SYN / SYN ACK
CLOSE / -
SYN SENT
SYN RCVD
SYN / ACK
ESTABLISHED
ACK / -
SYN ACK / ACK
19
TCP Disconnection
true
FIN / --
ESTABLISHED
auto- close
false
CLOSE / FIN
-- / FIN ACK
-- / ACK
CLOSING
CLOSE WAIT
FIN WAIT 1
FIN / ACK
ACK / -
CLOSE / FIN
ACK / -
FIN ACK / ACK
TIMED WAIT
LAST ACK
FIN WAIT 2
FIN / ACK
timeout / -
CLOSED
ACK / -
20
TCP Flow Control

• Sequence numbers correspond to octets within user
  data, instead of TCP segments.
• Window size field tells how many octets can be
  sent starting at the last byte acknowledged.
• Principle for flow control window once window
  space has been granted for specific octet
  numbers, the window space should not be revoked
  in a subsequent acknowledgement.
• Window size of 0 is legal octets up to
  acknowledgment number 1 have been received,
  but sender is requested not to send more data.
• Permission to restart indicated by sending a
  segment with the same acknowledgment number, but
  with a non-zero window size.
• TCP also uses a separate congestion window the
  sender may send only the number of octets
  determined from the minimum value of the two
  windows.

21
TCP transmission policy - 1
Sender
Receiver
Application writes 2K of data
0 4K
2K
Application writes 3K of data
2K
Application reads 2K of data
Sender blocked
2K
1K
22
TCP transmission policy - 2

• Window size of zero normally means that sender is
  blocked.
• Exceptions
• Sending of urgent data.
• May send 1 octet segment requesting receiver to
  send window size.
• Avoids deadlock when acknowledgment lost.
• Senders are not required to transmit data
  immediately on reception from application.
• Attempt to avoid sending segments with 20 octets
  of TCP header (and 20 more octets of IP header),
  and only 1 octet of user data.
• Acknowledgments are often delayed up to 500 ms,
  in the hope of collecting some user data to send
  along with it.

23
TCP transmission policy - 3

• Nagles algorithm for sparse user data (i.e. slow
  sender)
• Send first octet in a segment.
• Buffer all octets from application at sender,
  until first acknowledgment is received or buffer
  contains maximum segment size.
• Then, send entire buffer in one segment.
• Further data is buffered as above.
• Clarks algorithm for receiver slow to read data
• Receiver should not send a window update until
  either
• There is enough space in the buffer to handle the
  minimum segment size advertised in connection
  phase.
• The buffer is half empty.
• Avoids sending acknowledgments with 1 octet
  window size.

24
TCP Timer Management

• How long should the retransmission timer be?
• More difficult than in data link layer, since the
  round trip travel time variance is greater

T
Probability
10
20
30
20
40
60
Round trip time (msec)
Data link
Transport

• Too short unnecessary retransmissions that will
  clog network.
• Too long performance reduction due to long
  delay before retransmission.

25
Dynamic Timer Setting

• Algorithm due to Jacobson (1988).
• Keep variable RTT that is best current estimate
  of Round Trip Time to destination.
• Measure time M required for acknowledgment to
  return for a segment.
• Update RTT by
• RTT a RTT ( 1 a ) M
• where a determines how much weight to give old
  and new values.
• Typical value for a is 7 / 8
• Use mean deviation as an approximation to the
  standard deviation of RTT
• D b D ( 1 b ) RTT  M
• Set timeout to be
• timeout RTT 4 ? D

26
Additional TCP Timers

• Karns algorithm (for noisy, wireless media)
• Do not use retransmitted segments to update value
  for RTT .
• Instead, double timeout value until segments get
  through.
• TCP persistence timer
• Prevents deadlock when current window size is 0,
  and acknowledgment increasing window size is
  lost.
• When persistence timer expires, ask receiver for
  window size.
• TCP keep alive timer
• Timer expires after long interval with no
  messages.
• On expiry, send a message to receiver asking are
  you still there?
• Connection terminated if no response.
• TCP close timer ensures all packets die on
  connection termination.

27
UDP - User Data Protocol
Source Port
Destination Port
UDP length
UDP checksum
32 bits

• provides ability to send raw IP datagrams
  without establishing a connection
• described in RFC 768
• often used for client-server applications with
  one request, one reply
• UDP length number of bytes in datagram,
  including 8 byte header
• Check sum field provides error detection
• calculated in same manner as TCP, including
  pseudo-header
• 0 if not used

28
Sockets Network user interface

• Functions
• new Socket Create a new communication end
  point
• Bind Associate a socket with a specified port
  number.
• Listen Announce willingness to accept
  connections
• Queue size for incoming connection requests can
  be specified.
• Accept Block the caller until an incoming
  connection request arrives
• On receipt of a request, return.
• Connect Actively attempt to establish a
  connection
• Send Send some data over the connection
• Receive Receive some data from the connection
• Close Release the connection

29
Step 1 Create a socket
File descriptor or object reference

Our program
3
socket
Set of local ports
?
23
80
1237
Host
123.90.47.122
Network (IP) address
To network

• Creates attachment point from application to
  network.
• May be viewed by program as a file descriptor (as
  shown above) or an object reference.

30
Step 2 Bind the socket to a port
Our program
File descriptor
3
socket
Set of local ports
23
80
1237
Host
123.90.47.122
To network

• Chooses a port (which must be currently unbound).
• Required for servers client often bind
  implicitly to next available port while
  connecting.
• Port numbers from 0 to 1024 cannot be bound by
  application programs operating system or root
  privileges are required.

31
Step 3 (server) Listen for connections
Our program
File descriptor
3
socket
Set of local ports
23
80
1237
Host
Buffers for incoming connection requests
123.90.47.122
To network
32
Step 4 Incoming connection request
client program
server program
3
3
socket
socket
client host
1237
23
80
1237
SYN
123.90.47.122
191.36.42.1

• Client performs a connect operation
• Client needs to obtain network address of server.
• Call to Domain Name System (DNS) server used to
  obtain IP addresses from a domain name.

33
Step 5 Accept a connection request
our program
3
4
Socket for new connection
Socket for listening
host
80
123.90.47.122
To network

• A new socket is created for handling the data
  transfer.
• The original socket continues listening for
  connection requests.

34
Step 6 Create new process or thread
Server process/thread
Handler process/thread
3
4
host
80
123.90.47.122
To network

• Why?
• Server can handle multiple, simultaneous
  connection requests
• A separate handler process can be created for
  each request

35
Step 7 Data transfer phase
client program
handler
server
4
3
3
socket
client host
1237
23
80
1237
123.90.47.122
191.36.42.1

• Either side can send or receive data

36
Step 8 Close sockets
client program
handler
server
4
3
3
client host
1237
23
80
1237
123.90.47.122
191.36.42.1

• Sockets are independently closed at each side of
  the connection.
• Listening socket must also be closed
  independently if server shuts down.
• After waiting for twice the maximum packet
  lifetime, the port is released and is available
  for use again.

37
Socket attributes

• Local address (port network address) if bound
• Remote address (port network address) if
  connected
• Stream vs. datagram
• Usual value stream
• Protocol TCP, UDP, other
• Blocking vs. non-blocking
• Usual value blocking
• A blocking socket will stop and wait for the
  following operations to complete
• Accept wait until connection request arrives
• Receive wait until data is available

38
The Application View of a Socket

• In C, a program views a socket as a file
  descriptor.
• On Unix/Linux, socket( ) is a direct call to the
  operating system.
• Calling socket( ) returns a file descriptor.
• On Windows, you can use the basic interface
  directly via the winsock library.
• In C and Java, sockets are implemented within a
  class library, and then the programs view of a
  socket is an object reference.

39
Java Sockets - 1

• Primary classes
• Socket used for TCP outgoing connection or data
  transfer streams
• ServerSocket used to listen for incoming TCP
  connections.
• DatagramSocket used for UDP datagrams (if
  allowed by security manager)
• Instances of Socket and ServerSocket are always
  blocking.
• The Java class library uses constructor
  parameters to combine some of the steps.
• Creating a new Socket can also include the
  connect step.
• Creating a new ServerSocket can also include the
  bind, and listen steps.

40
Java Sockets - 2

• Java stream sockets have separate input and
  output streams.
• Typical usage
• Get input and/or output streams from socket
• Wrap stream in one of the stream classes.
• Use stream write or read methods to send or
  receive.
• Close streams and socket.
• All socket operations may throw IOExceptions (or
  more specific exceptions).

41
Java Sockets Example client

• // Implicit local port bind done by socket
  constructor
• socket new Socket( ?127.0.0.1?, 80 ) // remote
  IP port
• // Get and wrap input / output streams
• inStream new ObjectInputStream(aSocket.getInputS
  tream())
• outStream new ObjectOutputStream(aSocket.getOutp
  utStream())
• // Data transfer
• outStream.writeObject( object1 )
• object2 inStream.readObject( ) // Blocks until
  data arrives
• // Close streams and socket
• inStream.close()
• outStream.close() // Sends TCP 'FIN'
• socket.close() // Unbind socket