Transport Control Protocol

1
Transport Control Protocol

• TCP

2
Connection-Oriented Service

• Connection-oriented service has a handshake
  period
• During this time, a logical connection is made
  with the destination node
• The connection is logical, since all packets
  are forwarded individually, just like with UDP
• Typically, connection-oriented service provides
  reliability, meaning
• Acknowledgements are used to ensure packets
  arrive
• Checksums/CRCs are used to ensure data integrity

3
Transport Control Protocol (TCP)

• Like X.25, TCP provides connection-oriented
  delivery at a high level layer
• X.25 provides it at the Transport OSI layer
• TCP provides it at the Transport IP layer
• Providing connection-oriented delivery at a high
  level allows TCP to be applied to any network
• Thus the ability to use TCP/IP over Ethernet,
  Token Ring, etc.
• However, providing connection-orientation at a
  high level means that the network is not
  necessarily optimized for connection-oriented
  delivery
• For example, Ethernet is optimized for
  connectionless delivery

4
TCP

• The essence of TCP is to provide an apparently
  continuous stream of data
• Thus, above the transport layer
• Data is not fragmented (into packets)
• Data is in order
• Lost packets do not occur
• Thus, the transport layer (and layers below it)
  must handle
• Segmentation and reassembly (SAR)
• Acknowledgements

5
Segmentation vs. Fragmentation

• Segmentation is basically the same as
  fragmentation, with a few differences
• Fragmentation (IP layer)
• only occurs when transmitting a packet whose
  size is larger than the MTU of the destination
  network
• Any router (connecting two different network
  types) could theoretically fragment packets
• Fragmentation can almost be considered an
  emergency practice (what to do when something
  goes wrong)
• Segmentation (TCP layer)
• Occurs for all data streams, to divide the data
  into packets (above TCP layer data is continuous)
• Only the source host will segment packets
• Segmentation is a normal part of TCPs job

6
TCP

• TCP is a reliable protocol
• All data sent through TCP is automatically
  divided into packets
• Each of these packets is ensured to be sent by
  requiring the destination acknowledge the packets
  when they are received
• The destination, knowing it will eventually
  receive all messages, only has to reorder those
  messages into an apparently continuous stream of
  data flow

7
TCP Stream Delivery Protocol

• TCP abstracts data communication to appear as an
  apparent stream of flowing data
• The source sends data as a stream into the
  network
• The destination node receives data from the
  network in an identical form
• The data arrives in the same order as it was sent
• All data sent, arrives (in its proper position)
• This is known as stream orientation, a format
  where the data is oriented in such a way as to
  appear as a direct stream from source to
  destination
• In reality, however, the data is sent as packets
  (using IP datagrams, for example)

8
TCP Stream Delivery Protocol

• TCP is normally achieved by using buffering
• Data is collected from the stream (and stored
  into memory), until a certain amount has been
  obtained
• This data is packaged into one or more network
  packets (e.g. IP datagrams) and sent to the
  destination using connectionless delivery
• The destination should send an acknowledgement
  back to the source
• If this acknowledgement fails to arrive after a
  specified length of time, the source will
  retransmit the packet
• The destination node buffers the incoming packets
  into memory, where they can be read (byte by byte)

9
TCP Stream Delivery Protocol

• The TCP/IP service layers do not contain a
  Presentation layer
• For this reason, both UDP and TCP require that
  the applications recognize their own data formats
• For example, using TCP to connect and send an
  E-Mail message can be achieved using the existing
  data format (or language) known as SMTP (Simple
  Mail Transfer Protocol)
• Using TCP to connect and request WWW pages can be
  achieved using HTTP (HyperText Transfer Protocol)
• These protocols are implemented (generated and
  recognized) inside the applications themselves

10
TCP Stream Delivery Protocol

• Since TCP is actually implemented using packets
  (e.g. IP datagrams), it was possible for TCP to
  ensure bi-directional communication across its
  connections
• Transfer across TCP streams is full duplex

11
Connection Establishment

• TCP uses a three-way handshake to establish a
  connection
• This means 3 messages are exchanged before a
  connection exists
• The first message (SYN), sent by the machine
  issuing the active open request (A), is a
  request for connection to the destination (B)
• The second message (SYN/ACK), both an
  acknowledgement of the first message as well as a
  request for connection to A, is sent by B
• The third message (ACK) is an acknowledgement to
  B (from A) for the second message

12
TCP Handshake
SYN
Connection Establishment (Handshake)
SYN/ACK
ACK
Transmission of data
13
TCP Reliability

• TCP provides reliability by requiring recipient
  nodes to send acknowledgments
• Acknowledgements are sometimes called ACKs
• When a packet is received by the destination, an
  ACK is sent back to the source
• When the source receives the ACK, it sends the
  next packet
• And so on, and so on,

14
TCP Reliability
S
D
Network
M
M
M
M
M
M
M
A
A
A
A
A
A
A
M
M
M
M
M
M
M
A
A
15
TCP Reliability

• If a packet is sent, and no ACK is received
  within a certain time, the message will be
  retransmitted
• This time is called the timeout
• It is possible that the original packet was
  received, but the ACK was somehow lost
• TCP networks treat both situations identically
• The destination will receive the packet again,
  ignore it (it already has the data), and
  acknowledge it again
• Hopefully this time, the acknowledgement will be
  received

16
TCP Reliability
S
D
Network
M
M
M
M
M
M
M
5
6
7
8
9
10
17
TCP Reliability

• If each node waited for acknowledgements without
  transmitting data, it would involve wasteful
  delays between packets in a series
• TCP uses a scheme called the sliding window
  technique to solve this problem

18
Sliding Window Technique

• The sliding window technique allows a
  transmitting node to transmit more than one
  packet without waiting for an ACK
• Nodes cannot transmit more than S packets beyond
  the first unacknowledged packet
• S is known as the window size
• Thus, transmitting nodes have a window of up to
  S packets, all of which have already been sent
• Some of these packets may be acknowledged
• At least the first packet is unacknowledged (but
  sent)
• When an ACK is received for the first packet in
  the window, another packet can be sent
• The window index can be increased by one

19
Sliding Window Technique

• Lets see an example without a sliding window
• Packet size 4, Data abcd efgh ijkl mnop

SYN
SYN/ACK
ACK
1 abcd
Connection established
Passage of Time
ACK 1
2 efgh
ACK 2
3 ijkl
ACK 3
etc
4 mnop
20
Sliding Window Technique

• Lets see the same example with a sliding window
• Packet size 4, Window size 3, Data abcd efgh
  ijkl mnop

SYN
SYN/ACK
ACK
1 abcd
Connection established
2 efgh
3 ijkl
Passage of Time
ACK 1
ACK 2
ACK 3
4 mnop
ACK 4
21
Piggybacked ACKs

• Often two node communicate back and forth
• When an acknowledgement is to be sent from A to
  B, as well as a data packet, the ACK can be added
  to the packet and sent to B as one packet
• Essentially, only the sequencing number is
  required to indicate that a message has been
  received
• Sequencing numbers are discussed later
• Rather than send a small packet (ACK) followed by
  a larger packet (data), the node sends a single
  larger packet (data with piggybacked
  acknowledgement)

22
TCP Layers
Application
Transport Control Protocol (TCP)
Internet Protocol (IP)
Network Interface Hardware
23
TCP/UDP Layers
Application
TCP
UDP
Internet Protocol (IP)
Network Interface Hardware
24
TCP Header Information

• As stated previously, TCP is built on top of IP
  datagrams
• These datagrams must arrive correctly
• Therefore, TCP streams are often created using
  the same information as would be used in the IP
  datagram header
• Address (network and machine portion)
• Header checksum
• etc,

25
Ports

• Multiple TCP streams can be active on any machine
• Therefore, ports should be used to represent
  which stream is which
• These ports are the same ones used for UDP
• This makes sense, considering both use IP
  datagrams for their implementation

26
Sockets

• Sockets, to programmers, represent connections to
  the network
• In some sense, a socket are associated with a
  network port on the machine
• A machine (and even a single program) may have
  several open sockets at any time

27
Sockets

• In UDP, sockets can be shared
• Datagrams from different destinations can be
  received on the same socket
• Datagrams can be sent to multiple destinations
  through the same socket
• In TCP, sockets can not be shared
• TCP sockets (or stream sockets) represent an
  active connection with the other side
• Both source and destination must have an active
  socket open for communication to occur

28
Stream Sockets

• With stream sockets, one side must initiate the
  connection
• The side that will accept a connection requests a
  passive open with its operating system
• This indicates that the OS should accept incoming
  connection requests
• A port is associated with the passive open, and
  can be used by the initiating node when
  requesting the connection
• The side that initiates a connection requests an
  active open
• The initiating node requests a connection with a
  given machine (specified by its address) at a
  particular port
• If the machine has a passive open registered at
  that port, the connection will be accepted,
  otherwise it will not

29
Stream Sockets

• The socket that represents a passive open
• Is called a server socket
• Represents the willingness to accept connections
• The socket that represents an active open
• Is called a client socket
• Represents the act of actually connecting to a
  server socket

30
Data Corruption

• TCP uses a 16 octet checksum to ensure that data
  has not been corrupted
• If data is changed in any way, the checksum
  computed using the data at the destination will
  be different than the checksum computed on the
  source side (and transmitted along with the data)
• If checksums do not match
• Data is corrupt
• The checksum is corrupt
• Both situations are treated identically in TCP,
  data is retransmitted

31
Retransmission

• Similar to re-collision avoidance backoff,
  unacknowledged packets are sent after increasing
  timeouts
• This prevents packets from being indefinitely
  lost because the timeout value is too short for
  extremely high network usage situations
• Unless a message is undeliverable, in any amount
  of time, the message will eventually reach its
  destination and be acknowledged

32
Connection Use

• Once a connection has been made, sequence numbers
  are used to represent packets that make up the
  data stream
• Sequence numbers indicate the position of the
  data in the packet in the data stream