The Transport Layer
This layer, layer 4, accepts arbitrary data from the next, the session layer, and arranges it into packets suitable for the network layer (packetization). Similarly, it receives packets from the network layer and reconfigures them in the correct order for the session layer (depacketization).
This layer can manage network connections, maybe sending one data stream out over several connections to improve throughput, or multiplexing several data streams over one connection to save money. This layer also provides the type of service available to the user: examples are reliable (error-free), order preserving, connection oriented or connectionless.
It would be natural to want all our data transmissions to be 100% perfect: the bits that arrive are exactly the bits that were sent. However, arranging this can be very difficult given the unpredictable nature of hardware. Techniques (e.g., acknowledgement frames) can be used to approach reliability, but there is a cost (e.g., an acknowledgement frame can reduce the time and space that is available for real data). Sometimes we would rather not pay the cost, but instead allow for a margin of error in the data: transmission of audio is such a case, where slightly incorrect data is fine, but delayed data is not. In other cases (e.g., payroll data) we are happy to pay the overhead to get the 100% reliability.
Similarly, in a packet-oriented system, it may or may not be important that the packets containing the data arrive in the exact same order they were sent. Imposing order may cause extra expense that you might prefer not to pay in some applications.
A connection-oriented network is one where a path is made from the source to a destination and all data flows along this path. For example, when making a telephone call, a connection is set up before the data (the speech) can flow. In pre-digital days, telephone exchanges used to set up a physical copper path from caller to callee. A connectionoriented system is best when there needs to be good, smooth, uninterrupted flow of data.
In a connectionless network no connection is made and each packet is treated individually. This is like the postal system, where each letter is delivered individually. Two letters from the same source to the same destination could quite easily go via different routes and it is very possible that a later letter could be delivered before a letter posted earlier. A connectionless system is best when the data is small or irregular and you do not want the overhead of setting up a connection. Connection oriented is normally associated with circuit switching and connectionless with packet switching.
TCP/IP is considered to be reliable, order preserving and connection oriented, though the connection path is more conceptual than real.
In this tutorial:
- Layering Models
- The Seven Layer Model
- The Physical Layer
- The Data Link Layer
- The Network Layer
- The Transport Layer
- The Session Layer
- The Presentation Layer
- The Application Layer
- How the Layers Fit Together
- Why Layers and Encapsulation?
- The Internet Model
- The Link Layer
- Models and Protocols
- Comparing OSI and Internet Models