Home > Experiential Learning > Understanding Transport Layer Basics – Experiential Learning Series

## Authors

### Ram P. Rustagi

Professor, CSE Dept, KSIT

### Viraj Kumar

Professor, CAOS, IISc

### Abstract

The transport layer in a network stack provides end-to-end connectivity to the application layer. In the internet stack, there are two main transport layer implementations in use: TCP (which provides reliable service delivery) and UDP (which does not). Unfortunately, many people misconstrue these characterizations as “TCP provides guaranteed delivery” and “UDP may make errors”. This article attempts to clarify such misconceptions. It also discusses practical aspects of these two protocols (including TCP’s streaming service and UDP’s message-oriented service) and illustrates concepts with simple hands-on exercises.

### Introduction

The transport layer in the TCP/IP stack provides end-to-end connectivity between applications on two machines, which could be located anywhere in the world. For example, when a user accesses google.com in the browser, it establishes a reliable TCP (Transport Control Protocol) [1] connection between the user’s browser and one of Google’s servers. The TCP layer makes use of the underlying IP(Internet Protocol) [3] layer in the network and provides the required transport services to the application layer (e.g. HTTP [4]), in order to deliver usable data to the applications involved.

At its core, the Internet only provides the IP layer, which in turn only provides “best effort delivery”. It is important to note that “best effort delivery” does not provide any guarantee that a given packet will be delivered (or even that it will be delivered within a certain time-frame). It is quite possible that the IP layer loses some packets, and it does not retransmit such lost packets. Thus, IP does not provide any service guarantee with respect to lossless service. Further, since different IP-layer packets may traverse different paths between the same source and destination (on account of routing changes and updates in the IP network), these packets may be delivered out-of-order at the destination. Similarly, the IP layer does not guarantee that delivered packets will not be corrupted. Error detection at the IP layer is limited only to IP packet headers, whereas the data in the packet may get corrupted. Thus, a useful analogy for “best effort delivery” is postal delivery to a difficult-to-reach location. While the postal service promises its best effort, mail to a recipient may get lost or damaged on the way, and letters to the same address may arrive out of order.

Thus, even though the IP layer provides best effort delivery of packets given to it, it is the responsibility of the transport layer to provide some sort of reliability while depending upon underlying unreliable IP layer. Once again, the postal analogy is helpful. If the postal service discovers that a letter cannot be delivered due to an incorrect destination address, it tries to return the letter to the sender. Similarly, the receiver may confirm receipt of the letter via a letter in reply.

The transport layer helps two applications maintain the illusion that they are directly connected (a logical connection), even though the physical connection between them may consist of a number of intermediate network devices such as switches, routers, firewalls, etc. In any communication, the sender application breaks the data into segments (known as packets at the transport layer). The transport layer is responsible for delivering these packets to the receiver application, as per expected service delivery.

The two most commonly used transport layer protocol implementations for the Internet are transport Control Protocol (TCP) and User Datagram Protocol (UDP). Both of these differ significantly in the way they provide data delivery to applications.

UDP provides unreliable, connectionless and message-oriented service to applications that use it. In contrast, TCP provides reliable,connection-oriented, streaming, and in-order delivery service to applications that use it. When an application hands over transport layer packets to UDP, it is generally called a datagram. (At times, this term is also used for packets at the network-layer Internet Protocol, or IP). Similarly, a transport layer packet to TCP is generally referred to as a segment.

We will discuss each of the italicized terms above, as well as packet corruption and error control. For experiential understanding of these concepts, we will use the nc (Netcat) tool [5] and some simple Python programs. The Python programs for both the sender and receiver using both TCP and UDP along with instructions for their usage are available at [6]: udp_client.py, udp_server.py, tcp_client.py, and tcp_server.py.

The basic setup for the exercise that follow involve two systems (laptops, desktops, etc.) connected via a network consisting of two simple (unmanaged) switches, as shown in Figure 1. This is the simplest possible representation of the internet/network that allows us to introduce disturbances (such as network breakdown) without the applications being directly aware of them. The applications rely only on the transport layer protocol that provides the end-to-end logical connection. The internet connectivity shown here is not mandatory, but it represents a typical setup where users would like to remain connected to the internet while carrying out these experiential learning exercises.

Figure 1: Basic connectivity

An application at the transport layer needs transport layer addresses (port numbers). The Server program would run on a known (published) port number, which the Client program needs to be known. By default, the port for the Server is taken as 9999, but this can be changed using the option -p for both the TCP and UDP versions of the Server program.

Pages ( 1 of 5 ): 1 23 ... 5Next »