[Back to Lecture Notes page]

The Transport Layer

Topic Outline:

• Services of the Transport Layer

• The Internet Transport Protocols

• Transport Control Protocol (TCP)

• User Datagram Protocol (UDP)

• Network Performance Issues

Transport Layer Services

• Purpose: To provide reliable, efficient transport of data from a source to a destination across a network for the users using applications.

• Compare with DLL, which only deals with adjacent machines (source to next router, router to next router, etc).

• Why have a Transport layer (TL)?

• Why not just depend on the NL and DLL services to provide the service?

• Because NL and DLL are services of the subnet, which is beyond the control of the source and destination.

• We want a layer existing in the source and destination hosts which is able to handle problems like errors and delays.

Some Terminology

• Transport Entity: the hardware and software which provides TL services

• Transport Protocol Data Unit (TPDU): the message sent by the TL of the source to the TL of the destination.

• The primary aim of TL is to enhance the QoS (Quality of Service) provided by the NL.

• If the NL already provides very good QoS, then the job of the TL becomes very simple.

• The TL allows the applications to specify what level of QoS required, and tries to achieve it based on the available NL services.

Transport Layer Service Primitives

• Service primitives applications can call on:

• LISTEN
• CONNECT
• SEND
• RECEIVE
• DISCONNECT

• How it works: Fig 6-5 p486

Elements of Transport Layer

• The purpose of TL is very similar to DLL, Eg:

• Error Control

• Sequencing

• Flow Control

• But because TL works over a subnet rather than between adjacent machines:

• We need addressing

• Establishing and releasing a connection more complicated

• Subnet may delay a packet

• The TL needs to handle a much larger number of connections.

Establishing a Connection

• Complicated because of delayed and duplicated packets.

• Eg. Hosts sends packets to establish connection, transfer data and release connection, but all the packets are duplicated, and received a second time.

• Possible solution: keep track of obsolete connection numbers

• Requires hosts to keep history of connection numbers indefinitely

• Cannot use algorithms like sliding windows because we don’t know how long the duplicates can be delayed for.

Establishing a Connection

• Another solution: ensure packets are not held in the subnet indefinitely, by

• Restricted subnet design

• Hop count for the packets

• Time-stamping the packets

• But having the packets’ lifetime restricted, we can use sliding-window-like algorithms

Releasing a Connection

• Asymmetric release: like telephone connection, when one side hangs up, the connection is released

• Symmetric release: connection is two unidirectional connections, must be released separately.

• Asymmetric release can result in data loss. Eg. Fig 6-12 p498.

• Better to use symmetric release

Symmetric Release

• How it works:

• One party sends a "I am done" message to another.

• The other party sends an acknowledgment.

• Both parties release the connection.

• Two-army problem: can’t ever guarantee acknowledgment is received.

• How about sending an acknowledgment for the acknowledgment? - When do you stop?

• It can be proven that no protocol is guaranteed to work for all scenarios.

[Back to Lecture Notes page]