Chapter 2: Layers of the Internet

Four layers of the Internet:

1. Application Layer – This is the topmost layer of the Internet. Applications like a browser use this layer to send commands like ‘Get’ requests to the server. This is the layer where the request begins.

2. Transport Layer – Adds TCP header (explained below, for now just assume that this is some kind of information added by this layer) to the request

3. Network Layer – Adds IP header (main part of IP header are the source and target IP addresses) to the request. The resulting object is called ‘IP datagram’

4. Link Layer – Adds Ethernet header to the request. The resulting object is called ‘Ethernet frame’

The above 4 layers form a complete block that is sent from one physical–link to another.

Each physical–link/router strips the previous Ethernet header, checks the destination added by network layer and resends the data adding its own Ethernet header.

So, with every hop on the router, source Ethernet address changes to the current router. IP address of the source and the target remain the same.

TTL (Time-To-Live)

Every datagram has a field called the TTL. This field determines the lifetime of a datagram during its hop from one router to another. Every router decrements this field by one and the datagram expires when the TTL reaches zero.

Routers and Switches: Address lookup tables

Routers may have switches to form the connection between them. These switches only look at the Ethernet addresses to decide what switch/router to pass the information to.

Routers have a built-in lookup that tells them which packet should be sent to which router. It’s kind of a pattern matching. For example, a router may have the following map:

IP Address	Destination Router
122.254.x.x	1
290.x.x.x	2
10.141.141.x	3
x.x.x.x (default)	4

Advantage of Packet Switching

This approach really has a very big advantage when a packet has to find its path in a huge network. Imagine the conventional graph theory method of finding a path between two nodes. If the same approach were to be followed here, whole of the Internet would remain occupied only in finding paths among billions of routers and switches.

1. But in the above approach, every packet is self-contained at all the intermediate routers.

2. Also, this approach provides for a very efficient way to share links/paths between 2 nodes. For example, if path P is used for communication between A and B, the same path can be used to diverge some of the traffic from say X to Y. Even though, path P may not form the shortest path between X and Y, there is nothing stopping us to add some extra nodes while hopping from X to Y and this provides for a very good sharing of nodes/paths thus improving efficiency and avoiding network clogging of some paths.

3. This approach also allows transferring data parallelly from A to B, again improving the performance as compared to a dedicated connection between A and B.

4. It is also more robust because even if one path fails, transmission can be done through other paths. Suppose a message was intended to be sent from point A to B using nodes C, D and E. The address lookup tables of C, D and E are configured in such a way that the message accurately hops from one node to another and reaches its destination. If now, another node needs to be added to this path, it can be added conveniently between any 2 nodes and only one node's address tables would need to be updated. Also, if some node stops functioning, intelligence could be added to the nodes to look for an alternate path and update their address tables. Now compare this with the alternate approach of finding the entire path first from A to B, encoding this information in the headers and then sending the packet. That approach would simply not work in this case where nodes are being added/dropped dynamically into the system.

5. The link between 2 nodes is used only when required. This is in contrast to a dedicated link which may remain idle for most of the time because those 2 nodes may not be using it.

Statistical Multiplexing

Packet Switching provides for ‘Statistical Multiplexing’. To understand this, consider a multiplexer having 2 input lines and one output. It multiplexes the inputs equally such the output is 121212...

Now, if suppose line 1 was to go blank for some time, we would get holes in output corresponding to the 1s. However, in statistical multiplexing, the output is never blank unless all input is blank. The term is called ‘Statistical Multiplexing’ because it automatically gives higher preference to an input line with statistically higher number of packets.

OSI Model with 7 layers

In 1980s, networking had another model which had 7 layers. But now that is mostly obsolete.

However, the layers between the two are connected as shown below:

Features of the IP Layer (Network Layer)

1. Hop-Routing: Each Datagram packet goes by hopping from router to router and all packets are not required to take the same path even if they have the same source and the destination.

2. Unreliable: IP Layer does not make any guarantee that the packets will be delivered. For example, if a router in the middle fails because its dropped outside of the network, the packets its holding will be lost and the IP layer cannot retrieve them.

3. Datagrams may arrive out of sequence: This follows directly from #1. If every datagram can arrive by a different route, there is a high probability that they can arrive out of sequence. It’s the responsibility of the higher layer to put them in sequence.