Transport Layer

Peter Wood

TCP/IP layers

TCP/IP 5-layer reference model

TCP/IP layers with some protocols

Internet protocols

Transport Protocols

Multiplexing and Demultiplexing

Transport-layer multiplexing and demultiplexing

Endpoint Identification

Transport-layer segment showing port-number fields

Well-known port numbers

The User Datagram Protocol

The Connectionless Paradigm

  • UDP uses a connectionless communication setup
  • a process using UDP does not need to establish a connection before sending data (unlike TCP)
  • when two processes stop communicating there are no, additional, control messages (unlike TCP)
  • communication consists only of the data segments themselves

Message-Oriented Interface

  • UDP provides a message-oriented interface
  • each message is sent as a single UDP segment, which means that data boundaries are preserved
  • however, this also means that the maximum size of a UDP message depends on the maximum size of an IP datagram
  • allowing large UDP segments can cause problems
  • processes sending large segments can result in IP fragmentation, quite often on the sending computer
  • UDP offers the same best-effort delivery as IP
  • this means that segments can be lost, duplicated, or corrupted in transit
  • this is why UDP is suitable for applications such as voice or video that can tolerate delivery errors

Connectionless Multiplexing and Demultiplexing

  • say a process on Host A, with port number 19157, wants to send data to a process with UDP port 46428 on Host B
  • transport layer in Host A creates a segment containing source port, destination port, and data
  • passes it to the network layer in Host A
  • transport layer in Host B examines destination port number and delivers segment to socket identified by port 46428
  • note: a UDP socket is fully identified by a two-tuple consisting of
    • a destination IP address
    • a destination port number
  • source port number from Host A is used at Host B as "return address":

    Inversion of source and destination port numbers

UDP Segment Structure

  • UDP segment is sometimes called a user datagram
  • it consists of an 8-byte header followed by the application data (sometimes called payload), as shown below

    UDP segment structure

  • Source port # identifies the UDP process which sent the segment
  • Dest port # identifies the UDP process which will handle the application data
  • Length specifies the length of the segment, including the header, in bytes
  • Checksum is optional (see below)

UDP Header Example (DNS Request)

DNS request packet capture

UDP Header Example (DNS Response)

DNS response packet capture

Internet Checksum

  • both UDP and TCP use a 16-bit Checksum field
  • the sender can choose to compute a checksum or set the field to zero
  • the receiver only verifies the checksum if the value is non-zero
  • note that the checksum is computed using ones-complement arithmetic, so a computed zero value is stored as all-ones

Checksum Example

Example checksum

  • to compute the checksum, the sender treats the data as a sequence of binary integers and computes their sum, as illustrated above
  • each pair of characters is treated as a 16-bit integer
  • if the sum overflows 16 bits, the carry bits are added to the total
  • the advantage of such checksums is their size and ease of computation
  • addition requires very little computation and the cost of sending an additional 16-bits is negligible

Example of Checksum Failure

Checksum failure

  • checksums do not detect all common errors, as illustrated above
  • a transmission error has inverted the second bit in each of the four data items, yet the checksums are identical

UDP Encapsulation

  • recall that each layer in the protocol stack adds its own header
  • each UDP segment is encapsulated in a network-layer (IP) datagram
  • each IP datagram is encapsulated in a link-layer frame

UDP Encapsulation

Protocols Using UDP

  • UDP is especially useful in client-server situations, when a client sends a short request to the server and expects a short response
  • if either the request or response is lost, the client times out and tries again
  • if all is well, only two packets are required
  • an example of an application that uses UDP in this way is the Domain Name System (DNS)

Transmission Control Protocol (TCP)

  • the Transmission Control Protocol (TCP) is the transport level protocol that provides reliability in the TCP/IP protocol suite
  • from an application program's perspective, TCP offers:
    • connection-oriented: an application requests a connection, and then uses it for data transfer
    • point-to-point communication: each TCP connection has exactly two end points
    • reliability: TCP guarantees that the data sent across the connection will be delivered exactly as sent, without missing or duplicate data
    • full-duplex connection: a TCP connection allows data to flow in both directions at any time
    • stream interface: TCP allows an application to send a continuous stream of bytes across the connection
    • reliable startup: TCP requires that two applications must agree to the new connection before it is established
    • graceful shutdown: TCP guarantees to deliver all the data reliably before closing the connection

End-To-End Service

  • TCP uses IP to carry messages, known as segments
  • each TCP segment is encapsulated in an IP datagram and sent across the Internet
  • TCP treats IP as a packet communication system:

    Example internet

  • as illustrated, TCP software is required at both ends of the virtual connection, but not on intermediate routers
  • from TCP's point of view, the entire Internet is a communication system capable of accepting and delivering messages without changing or altering their contents

Connection-Oriented Multiplexing and Demultiplexing

  • each TCP connection has exactly two end-points
  • this means that two arriving TCP segments with different source IP addresses or source port numbers will be directed to two different sockets, even if they have the same destination port number
  • so a TCP socket is identified by a four-tuple:
    (source IP address, source port #, destination IP address, destination port #)
  • recall UDP uses only (destination IP address, destination port #)

Multiplexing and Demultiplexing Example

  • an example where clients A and C both communicate with B on port 80:

Two clients communicating with the same port numbers

Reliable Data Transfer

  • TCP is a reliable data transfer protocol
  • implemented on top of an unreliable network layer (IP)
  • some problems:
    • bits in a packet may be corrupted
    • packets can be lost by the underlying network
  • some solutions:
    • acknowledgements (ACKs) can be used to indicate packet received correctly
    • a countdown timer can be used to detect packet loss
    • packet retransmission can be used for lost packets

Simple Reliable Data Transfer

  • a simple reliable data transfer protocol might
    • send a packet
    • wait until it is sure the receiver has received it correctly
  • such a protocol is known as a stop-and-wait protocol
  • performance of such a protocol on the Internet would be poor

Pipelined Reliable Data Transfer

  • a pipelined protocol allows for multiple data packets to be sent while waiting for acknowledgements
  • this results in better network utilisation
  • sender and receiver now need buffers to hold multiple packets
  • packets need sequence numbers in order to identify them
  • an acknowledgement needs to refer to corresponding sequence number
  • retransmission can give rise to duplicate packets
  • sequence numbers in packets allow receiver to detect duplicates

Packet Loss and Retransmission

  • TCP copes with the loss of packets using retransmission
  • when TCP data arrives, an acknowledgement is sent back to the sender
  • when TCP data is sent, a timer is started
  • if the timer expires before an acknowledgement arrives, TCP retransmits the data

Packet Loss and Retransmission - Example

Example retransmission

  • host on the left is sending data; host on the right is receiving it
  • TCP must be ready to retransmit any packet that is lost
  • how long should TCP wait?
  • the TCP software does not know whether it is using
    • a local area network (acknowledgements within a few milliseconds) or
    • a long-distance satellite connection (acknowledgements within a few seconds)

Adaptive Retransmission

  • TCP estimates the round-trip delay for each active connection
  • for each connection, TCP generates a sequence of round-trip estimates and produces a weighted average (mean)
  • it also maintains an estimate of the variance
  • it then uses a linear combination of the estimated mean and variance as the value of the timeout

Adaptive Retransmission - Example

Timeout and retransmission

  • the connection on the left above has a relatively long round-trip delay
  • the connection on the right above has a shorter round-trip delay
  • the goal is to wait long enough to decide that a packet was lost, without waiting longer than necessary
  • when delays start to vary, TCP adjusts the timeout to a value greater than the mean to accommodate peaks

TCP Segment Structure

TCP segment structure

  • Source port #, Dest port # and Internet checksum are as for UDP
  • Sequence number (32 bits) and Acknowledgement number (32 bits) are used to implement reliable transfer (see below)
  • Header length (4 bits) is the header length (including possible options) in 32-bit words
  • the flag field contains 6 1-bit flags (see below)
  • Receive window identifies how much buffer space is available for incoming data (used for flow control)

TCP Flags

  • URG flag indicates that the sender has marked some data as urgent
  • in this case, the Urgent data pointer contains an offset into the TCP data stream marking the last byte of urgent data
  • ACK flag indicates that the acknowledgement number field is valid (i.e. the segment is an acknowledgement)
  • PSH flag indicates that should be delivered immediately (PUSHed) and not buffered
  • RST flag is used to reset a connection, i.e. a confused or refused connection
  • SYN flag is used to establish a connection (see below)
  • FIN flag is used to terminate a connection (see below)

TCP Example (HTTP Request)

TCP segment for HTTP GET

TCP Example (HTTP Response)

TCP segment for HTTP response

Sequence and Acknowledgement Numbers

  • TCP views data as an ordered stream of bytes
  • sequence numbers are with respect to the stream of transmitted bytes
  • the sequence number for a segment is therefore the byte-stream number of the first data byte in the segment
  • the receiver uses the sequence number to re-order segments arriving out of order and to compute an acknowledgement number
  • an acknowledgement number identifies the sequence number of the incoming data that the receiver expects next
  • suppose Host A has received bytes 0 through 535 and 900 through 1000 from Host B, but not bytes 536 through 899
  • A's next segment to B will contain 536 in the acknowledgement number field
  • TCP only acknowledges bytes up to the first missing byte in the stream
  • TCP is said to provide cumulative acknowledgements

Example: Lost Acknowledgement

Retransmission due to lost acknowledgement

  • Host A sends one segment to Host B
  • this segment has sequence number 92 and contains 8 bytes of data
  • the acknowledgement from B is lost
  • A retransmits after its timer expires

Example: Single Retransmission

Second segment not retransmitted

  • Host A sends two segments back to back to Host B
  • acknowledgements from B arrive only after timeout
  • if acknowledgement for second segment arrives before the new timeout, the second segment will not be retransmitted

Example: No Retransmission Necessary

No retransmission necessary

  • Host A sends two segments back to back to Host B (as in previous example)
  • suppose the acknowledgement for the first segment is lost
  • if second acknowledgement arrives before timeout, A does not retransmit either segment

Flow Control

  • TCP uses a window mechanism to control the flow of data
  • when a connection is established, each end of the connection allocates a buffer to hold incoming data, and sends the size of the buffer to the other end
  • as data arrives, the receiver sends acknowledgements together with the amount of buffer space available called a window advertisement
  • if the receiving application can read data as quickly as it arrives, the receiver will send a positive window advertisement with each acknowledgement
  • however, if the sender is faster than the receiver, incoming data will eventually fill the receiver's buffer, causing the receiver to advertise a zero window
  • a sender that receives a zero window advertisement must stop sending until it receives a positive window advertisement

Flow Control Example

TCP flow control

  • sender is using a maximum segment size of 1000 bytes
  • receiver advertises an initial window size of 2500 bytes
  • sender transmits three segments (two containing 1000 bytes and one containing 500 bytes); then waits for an acknowledgement
  • the first three segments fill the receiver's buffer faster than the receiving application can consume the data, so the advertised window reaches zero
  • after the application reads 2000 bytes, the receiving TCP sends an additional acknowledgement advertising a window of 2000 bytes
  • sender responds by sending two 1000-byte segments resulting in another zero window
  • application reads 1000 bytes, so the receiving TCP sends an acknowledgement with a positive window size

TCP Connection Establishment

  • connections are established by means of a three-way handshake
  • each side sends a control message, specifying window size and Initial Sequence Number (ISN) which is randomly chosen
  • a random ISN reduces the chance of a "lost" segment from an already-terminated connection being considered part of this connection
  • the three steps are:
    • the sender sends a TCP segment (including window size and ISN) with the SYN flag on
    • the recipient sends a segment (including window size and ISN) with both SYN and ACK flags on
    • the sender replies with ACK

Example: Connection Establishment

TCP open connection

  • host 1 opens the connection with an ISN
  • host 2 accepts the connect request by sending a TCP segment which
    • acknowledges host 1's request (ACK flag on)
    • sets acknowledgement number to ISN+1
    • makes its own connection request (SYN flag on) with an ISN
  • host 1 acknowledges this request
  • note that the SYN flag "consumes" one byte of sequence space so that it can be acknowledged unambiguously

TCP Example SYN

TCP connection request


TCP acknowledgement and connection request

TCP Connection Release

TCP close connection

  • a three-way handshake is also used to terminate a connection
  • in this example, host 1 terminates the connection by transmitting a segment with the FIN flag set containing optional data
  • host 2 acknowledges this (the FIN flag also consumes one byte of sequence space) and sets its own FIN flag
  • the third and last segment contains host 1's acknowledgement of host 2's FIN flag

Congestion Control

  • packet loss typically results from buffer overflow in routers as the network becomes congested
  • congestion results from too many senders trying to send data at too high a rate
  • packet retransmission treats a symptom of congestion, but not the cause
  • to treat the cause, senders must be "throttled" (reduce their rate)
  • TCP implements a congestion control algorithm based on perceived congestion by the sender:
    • if it perceives little congestion, it increases its send rate
    • if it perceives there is congestion, it reduces its send rate
  • we will not cover the details of how TCP does this

Links to more information

See Chapter 3 of [Kurose and Ross], Chapters 25 and 26 of [Comer] and parts of Chapter 6 of [Tanenbaum].