PUCPR

2014

Introduction to

Computer Networks TCP/IP Layer Architecture

Edgard Jamhour

E N G L I S H S E M E S T E R

1. LAN: Local Area Networks: Ethernet Technology

Ethernet is a technology for implementing local area networks (LANs), and it is based on the

principle of physical broadcast.

The designation LAN (Local Area Networks) is used to describe a portion of a communication

network that uses a short range communication technology.

Currently, Ethernet is the most widespread technology to create LANs. The Ethernet

technology has evolved a lot in the last thirty years. When it was first introduced, it operated

in a non-switched mode, because all computers on the network shared a single medium. A

transmission in shared medium propagates to all computers on the network. This mode of

transmission is called (physical) broadcast.

Transmissions in Ethernet technology are made through structured messages called frames.

The header of a frame brings the destination addresses and source involved in the

transmission. The destination address indicates who should read the message, and the source

address identifies the transmitter.

As each computer in the LAN receives all messages, it must compare the destination address of

each received message with its own in order to determine whether the message needs to be

processed or not.

2. Ethernet II Frame

A frame is the smallest structure of information transmitted over a local network.

In a LAN, data is fragmented and transmitted in structures of limited size called frames. A

frame is composed of three parts: header, data and tail.

As will be explained later in the course, there are two variants in the format of Ethernet

(Ethernet II and IEEE 802.11) frames. The figure illustrates the format of Ethernet II frames. In

this format, the header consists of three fields: physical destination address, physical source

address and a code that indentifies the type of data transported. The size of the data field is

variable, and its maximum size is called Maximum Transmission Unit (MTU). The tail has a code

for error checking. It allows the receiver to detect, at some degree, if the received frame has

not been altered during transmission by noise or other interference.

The format of the physical address is defined by the Ethernet standards, and is usually called

MAC address (its meaning will be discussed later in the course). In this introduction, the

physical address will be represented by simple capital letters such as A, B, etc..

The Ethernet specification defines various aspects of the technology, such as the frame

structure, and the form of electrical or optical representation of bits. It also defines the

procedure for transmitting and receiving frames in a shared medium (Media Access Control –

MAC).

3. MAC: Media Access Control

In non-switched Ethernet, only one computer can transmit at a time.

In a non-switched Ethernet, only one computer can access the medium each time. If more than

one computer transmits at the same time, the data will be overwritten, and the frames will

arrive with errors at the destination.

In order to prevent two or more computers to transmit at the same time, the Ethernet

employs a decentralized technique of medium access control called CSMA / CD. The technique

is said to be decentralized because it consists of an algorithm that runs independently on each

of the computers on the network without the aid of a central entity. The acronym CSMA / CD

means: Carrier Sense Multiple Access with Collision Detection.

The CSMA / CD will be studied in more detail in the course, but for the moment, we will

describe its operation in a simplified form, as follows:

1) All computers must listen to the medium before transmitting;

2) If the medium is busy, the frames are stored in queues. When the medium is idle,

computers can transmit;

3) While transmitting a frame, the sender must continually compare the data sent with that

read from the bus. When two or more computers send frames simultaneously the bus voltage

levels observed in the bus differ from the logical representation of the bits of the original

frame, allowing the computer to detect the occurrence of a collision.

4) In the event of a collision, the frame must be transmitted again after a random waiting time.

5) The number of retransmissions attempts is limited.

4. Ethernet II Limitations: Propagation Delay

The propagation time between hosts affects the maximum performance of the network.

Over the years, the nominal speed Ethernet has increased considerably. According to its speed,

Ethernet received different names: Ethernet (10 Mbps), Fast Ethernet (100 Mbps), Gigabit

Ethernet (1000 Mbps), etc.. As we shall see, some of these speeds are not attainable in non-

switched mode.

Even if there is no collision, the non-switched Ethernet cannot reach its maximum rated speed,

because there is some loss of bandwidth every time the right to transmit is passed from one

computer to another.

This happens because the wave propagation speed of signal in the transmission medium is

limited. For example, consider a scenario where two computers, called A and B, are competing

for the right to access the medium. Suppose computer A gets the medium first. It puts a frame

on the bus at time t = to. The first bit of this frame will only be noticed by B at a time t = to + .

The time is the wave propagation time from A to B. For example, in the case of the electrical

signal, this wave propagation speed is about 200,000 km / sec. If computer B has a frame to

transmit, it can only do so at time t = to + + T. Thus, part of the capacity of the bus lost

forever.

The maximum occupancy (or efficiency) of an Ethernet bus depends on the average frame size

(T) and the propagation time of the signal between the most distant computers in the bus (),

and is given by:

Efficiency = T/(T+)

5. Ethernet II Limitations: Effects of the Distance

The maximum occupancy of the medium decreases with both: the distance between

computers and the transmission rate.

The transmission technology used in LANs suffers a large drop of performance when applied to

large distances.

The performance loss is more significant for small frames transmitted at high rates. For

example, let's consider an unfavorable scenario where the frame size is the minimum allowed

by Ethernet: 64 bytes.

For this frame size, the efficiency of a non-switched Ethernet operating at 10Mbps is 98%

when the maximum distance between computers is 200 m. When the transmission rate

increases to 100Mbps the maximum efficiency drops to 83.6%. At the same rate, but with a

distance of 800m, the efficiency is only 56%. This means that the effective throughput capacity

of the network is 56Mbps, and this capacity is shared by all the computers on the network.

For longer distances, the use of non-switched Ethernet technology becomes impractical

because the performance loss is very significant. Therefore, the maximum distance between

the computers in non-switched mode is limited by the standard.

One must observe that the power loss due to energy dissipation is also a limitation to the

maximum transmission distance over Ethernet. That’s the reason that the maximum length of

an Ethernet cooper wire is limited to 100 or 200 meters, depending on the type of the cable.

The energy dissipation can be reduced by using optical cables or the signal may be amplified to

compensate the energy loss using repeaters. However, the efficiency of the network will

remain limited regardless the type of cable used. It is a limitation that cannot be overcome

using the non-switched Ethernet.

6. Ethernet II Limitations: Collisions

Even with computers listening before transmitting medium, there is a possibility of collision.

The distance between computers on the non-switched Ethernet technology also affects the

possibility of collision.

To understand why this occurs, consider that two computers, A and B, share a bus and that a

signal placed in the bus takes seconds to go from A to B.

If computer A starts its communication at time t = t0, computer B only notices that the

medium is busy at time t = t0 + . If computer B initiates a transmission in the interval between

t0 and t0 + there will be a collision.

The probability of collision will increase with the distance between the computers (since the

value of increases proportionally with the distance).

A formula commonly used to predict the maximum performance of the switched Ethernet

network is not given by:

Efficiency = 1/(1 + 6,44 /T)

The expression assumes that all computers transmit following a Poisson distribution. The

expression represents the maximum achievable throughput of useful frames.

7. Ethernet II Limitations: Effects of Distance over Collisions

The distance between computers increases the collision probability

The distance between computers also significantly affects the performance of the non-

switched Ethernet due to the increased probability of collision. The loss in efficiency is more

significant in case of small frames. The figure illustrates the efficiency values for the worst case

scenario, where the size of the Ethernet frames correspond to the minimum value defined by

the standard. Another factor that affects the performance of the network is the transmission

rate. The collision probability is higher for higher rates.

In summary, we have observed that the non-switched Ethernet technology has limitations. The

number of computers is limited because only one computer can transmit at a time. Therefore,

the network performance decreases when many computers that are placed on the same bus.

The distance between the computers is also limited. To avoid collisions, computers listen to

the bus and transmit only if the bus is idle. The greater the distance between the computers,

the higher is the probability of collisions. Because the computers attempt to re-transmit

frames after collisions, the situation may quickly deteriorate if the computers are placed too

far away. There is a point where the network gets into a “collapse”, and remains in a state of

very low performance, where almost all frames are lost by collision.

As we shall see, to enable the operation of Ethernet at higher speeds and longer distances it is

necessary to interconnect computers using a device called Ethernet switch. A switch does not

use the principle of transmission by physical broadcast. In this case, the Ethernet network

receives the designation of "switched".

8. Hubs

Hubs or concentrators are devices which internally simulate the construction of a physical

bus.

The first evolution of the non-switched Ethernet happened with the introduction of network

devices called HUBS (or concentrators).

The introduction of HUBS enabled the replacement of coaxial cables by the UTP (unshielded

twisted pair) cables, used until nowadays.

Observe that HUBs offers no gains in terms of network performance, as they continue to

operate according to the principle of the physical broadcast. A frame received at one port of

the HUB is relayed to all other ports regardless the target address.

The gain achieved by the introduction of the HUB was an easier connectivity. In the coaxial

cable model, any opening of the bus involved a disruption of the network due to signal

reflection problems caused by impedance matching. At the HUB, you can insert and remove

computers without impairing the communication of other computers. The HUB also increases

the maximum distance between the computers because it works as a repeater (amplifier),

compensating losses in the signal level due to attenuation introduced by twisted pair when the

frame is retransmitted from one port to another.

9. Switches

The introduction of swiches changes the operation of the Ethernet to the “swiched” mode.

Ethernet switches are network equipments capable of forwarding an incoming frame only to

the port where the target computer is connected. To perform the forwarding operation, the

switch maintains in memory a table that indicates the address of the computers connected to

each one of its ports.

The process of filling the forwarding table is fully automatic. The Ethernet switch operates

transparently to the network computers. That is, it is not necessary to make any changes or

configuration on the computers so that they start operating with the switch. In fact, for

computers, intermediation taken by the switch is completely transparent.

As the figure shows, initially the routing table is empty. When computer A sends a frame to

computer C, the switch interprets the destination address and tries to find which port the

computer C is located. When a destination address is not located, the corresponding frame is

sent to all switch ports, i.e., the switch operates in a mode equivalent to a HUB.

However, every time a frame is received by the Switch, the source address of the frame is used

to update the forwarding table. In this case, the switch determines that computer A is

connected to the port 1. Accordingly, any response from the computer C to A is sent only to

port 1. Similarly, when frame from C is received by the switch, the switch learns the computer

C is connected on the port 3.

10. Collision Domains in Switches

Each switch port defines a collision domain. This is only possible collision between computers

connected to the same port.

Initially, Ethernet switches were very expensive, making impractical to connect only one

computer to each switch port. Thus, a common strategy was to connect multiple computers to

a single port, using HUBS.

Each switch port is an independent collision domain. It is only possible to have collisions

between computers connected to the same port. Internally, the switch features an array of

high-performance switching, which allows transmitting multiple flows in parallel.

As a computer only competes for the medium with computers connected to the same switch

port, the performance gain for the network is still very high.

Unlike hubs, switches allow to perform conversion of the transmission rate. For example, the

computer at port 3 may be a server operating at 1Gbps, while the computers operating on

port 1 and 2 are only 100 Mbps.

Note that the forwarding table of the switch can have multiple computers associated with a

single port entry.

11. Cascading Switches

Although significantly improve network performance, Switches still have limited scalability.

Currently, most Ethernet switches available in the market have 12 or 24 ports. To create larger

networks, you connect several switches together (cascading). There are several ways to

cascade switches. The manner indicated in the figure is one of the simplest.

Cascading switches permits to create large networks, but one cannot create a network of any

size using only this technology. Looking at the forwarding tables of the switches, we observe

that each switch must know the address of all computers in the network, even those

connected to other switches. This means that if we cascade 50 switches, and create a network

with 1000 computers, each one of the switches must have a forwarding table with 1000

addresses.

All switches cascaded together are still considered a single LAN. This definition is important

because the communication performed within a LAN follows the same principle, regardless the

fact that computers are connected to different switches.

Clearly, building a network of the size of the Internet (which currently has billions of devices

connected to) using only the Ethernet technology is completely unfeasible.

The Internet is organized using a different network topology called WAN (Wide Area Network).

In order to build WANs, it was necessary to introduce new network equipment (the routers)

and a new protocol (IP – Internet Protocol), that uses a different type of address.

12. WAN: Wide Area Networks

The WAN (Wide Area Networks) uses a different address scheme that allows you to connect

an unlimited number of switches in arbitrarily large distances.

In order to build larger networks, a network architecture called WAN (Wide Area Network) is

required. The Internet follows this architecture.

A WAN is formed by the interconnection of multiple LANs, using another type of network

device called a router. A router uses a very different address scheme than a switch. Instead of

mapping the individual addresses of computers to its ports, the router maps network

identifiers that represent LANs.

The tables used by routers are called "routing tables". The figure illustrates the structure of the

routing table of router 3. As we will see later, the routing table needs a few more columns of

information omitted from the figure.

The great advantage of this strategy is that the number of entries in the routing table is count

in networks, not computers. For example, the current Internet has 2.5 billions of users, but

"only" 500 000 networks.

Physical addresses defined by the Ethernet protocol do not have a network identifier. In order

to support the identification of the network another type of address is required: The IP

address.

13. Network Address: IP Address

IP addresses have 32 bits, i.e. 4 bytes, represented in dotted decimal notation.

The Internet Protocol (version 4) IPv4 protocol currently used on the Internet, uses 32-bit

addresses. A 32-bit address corresponds to four octets (bytes). In dotted decimal notation,

each octet is represented by a decimal number, calculated as if each octet was independent of

the others.

The figure illustrates the representation of the IP address 128.10.2.30. An IP address identifies

a computer, but also identifies the network the computer belongs. Indeed, the most significant

part of the IP address is the network identifier, and the least significant part is the host

identifier. The IP address of all computers in the same must have the same network identifier.

The number of bits that identifies the network is variable. In fact, the IP address alone has no

particular meaning. In order to interpret an IP address, it must be followed by another number

called "subnet mask". The subnet mask indicates how many bits are used to determine the

network identifier. The most compact way to inform the subnet mask is to use a / after the IP

address. For example:

128.10.2.30 / 8 means that the computer belongs to a network 128

128.10.2.30/16 means that the computer belongs to a network 128.10

128.10.2.30/24 means that the computer belongs to network 128.10.2

14. Packets

Packets are transported in the payload of frames

In computer networks each protocol defines a basic unit to transport data, named PDU

(Protocol Data Unit).

Ethernet and IP are different protocols. The PDU of Ethernet is called “frame” and the PDU of

IP is called “packet”.

As the Ethernet frame, an IP packet comprises a header and a data field (also called payload).

The header includes the source and destination addresses, and other information used

required to forward the packet through routers (omitted in the figure).

Importantly, the Ethernet and IP protocols work in a cooperative way. In fact, the packets are

transmitted in the payload of frames. The relationship is one-to-one, i.e., only one packet is

transported within a frame at a time.

The distinction between frame and packet is not always clear in the literature. Sometimes, it is

possible to find references to "Ethernet packets." In our course, however, the term frame is

used to designate the entire structure and the term packet only the IP-based structure, as

indicated in the figure.

15. Packets and Frames

Grouping computers into networks allows reducing the amount of information in the router's

table.

As we have seen, an IP address indentifies a computer and the network to which it belongs.

The IP addressing scheme determines that all computers in a same LAN MUST have the same

network identifier. Also, computers located in different LANs MUST have distinct network

identifiers.

A router can be considered as a computer with multiple network interfaces. Each interface

(port) of a router has different physical address. In a packet, the source and destination

addresses necessarily identify the end-to-end transmitter and the receiver of the package. In a

WAN, however, the physical addresses do not always identify a computer.

In fact, the physical addresses uniquely identify the entities involved in the part of the path

between two routers. This part of the path is usually referred to as “data link”. Say otherwise,

a “data link” is the part of the path that does not require jumping through a router. The action

of jumping through a router is referred to as a “hop”.

As indicated in the figure, when a packet is being transported between two routers, the

physical addresses indicate the routers interfaces of origin and destination. When the packet is

sent by the router to its final destination, the physical addresses indicate the router interface

and the computer receiving the message. The IP addresses, however, always indicates the end-

to-end source and destination.

16. Technology Independence

Packets are independent of the physical medium. Frames, however, must change to adapt to

the data-link technology.

The relationship between frame and packet also guarantees the independence of the IP

protocol regarding to the transmission technology used in the data links.

The figure above illustrates a scenario in which this concept is applied. A frame is created and

sent from LAN1 to LAN2. The LAN1 uses Ethernet Technology. Upon arriving at router 1, the IP

packet transported by the Ethernet frame is copied and inserted into a PPP frame, in order to

adapt to the transmission technology used in the data link connecting the two routers. PPP

(Point to Point Protocol) is a protocol used to transport information through serial lines, such

as phone lines. Upon arriving at router 2, the IP packet is extracted from the frame and placed

inside a Token-Ring frame.

In summary, one of the great advantages of the IP protocol is its independence of transmission

technologies. A LAN is formed by a single technology, such as Ethernet and Token-Ring. The

WAN, however, may be formed by heterogeneous transmission technologies.

17. Transport Protocols

Transport protocols, such as TCP and UDP, are responsible to deliver the payload of packets

to the process.

The IP and Ethernet protocols contain information that allows addressing a computer placed in

any LAN in a WAN. However, the computer is not the final destination of the communication,

but a process running in the computer.

The protocols responsible for addressing processes are called "transport protocols". Two

transport protocols are used in the IP technology: (Transmission Control Protocol) TCP and

UDP (User Datagram Protocol). TCP and UDP are never used simultaneously. Some

applications are based on TCP and others on UDP.

The reason for having two protocols is that they are designed to support applications with

different requirements. TCP is a reliable protocol, which offers several features to ensure the

delivery of messages, such as automatic retransmission of lost packets. TCP also interferes on

the rate that packets are sent. That is, the TCP imposes a transmission behavior to the

application. For the applications that this behavior is not suitable, the UDP protocol must be

used. The UDP protocol is quite light, leaving the transmission behavior to the application

level.

Both TCP and UDP use the concept of port numbers (16-bit integer numbers) to indentify

processes. Thus, when a process A in computer 1 sends a message to a process B in computer

2, it must include the port numbers in header of the message.

18. Port Numbers

Port numbers are addresses used by TCP and UDP. Well-known ports are mapped to the

standard applications defined by IANA (Internet Assigned Number Authority)

The IANA (Internet Assigned Number Authority) is responsible for defining how TCP and UPD

ports are mapped to applications.

A port addresses are 16-bit numbers, the port values range from 0 to 65535. IANA divide the

port numbers into three groups.

The first group is called well-known ports, and ranges from 0 to 1023. Generally, they are used

to address applications that are standard and vendor-independent, such as http (80), ssh (22),

SMPT (25) and telnet (23). To access the ports in this group it is necessary to have root

(administrator) privileges.

The second group is called registered ports, and ranges from 1024 to 49151. These port

numbers does not require root privileges. They represent vendor specific applications such as

database servers. For example, Oracle, SQL Server, MySQL are all database applications, but

uses different port numbers.

The third group is called dynamic or private ports, and ranges from 49152 to 65535. This range

is normally used by client applications. In a client-server communication, the client port is

random and the server port is fixed. The client ports are chosen by the operating system, so as

to avoid conflict with the ports of server or other client processes already running in the

computer.

19. Protocol Data Unit

Protocol Data Unit (PDU): Frame, Packet, Segment (TCP)/Datagram(UDP)

The unit of information carried by a protocol is generically called PDU (Protocol Data Unit). The

PDU of the Ethernet protocol is called frame and the PDU of the IP protocol is called packet.

The PDU of the transport protocols receives different names: segment for TCP and datagram

for UDP.

The PDU of transport protocols are placed in the payload field of IP packets, as show in the

figure. The header of the IP packet carries a field called "Protocol", which identifies whether

the PDU in the payload is the TCP or UDP (or any other) type.

Please, observe that we have omitted several protocol fields in the figure, as they are not

relevant to this introduction.

Thus, the Ethernet header precedes the IP header, which in turn precedes the header used by

TCP or UDP. This concept of placing the PDU of a protocol inside the payload of other protocol

is called a protocol stack.

Protocol stack is a main concept in packet switched networks such as the Internet. In a

communication, many protocols are used simultaneously. Each protocol is responsible for

providing a specific set of functions required by the communication. As we have seen, Ethernet

supply information for transporting a frame within a LAN, and IP across different LANs (i.e., a

WAN). TCP and UDP are responsible for delivering a PDU to a specific process running in a

operating system.

20. Application Protocol

Application protocols define a set of standard messages that permit clients and servers from

different vendors to communicate.

There are many different services implemented over the TCP/UDP/IP protocols. Typical

examples are email, web and remote access via SSH or TELNET. Many of these services follow a

client-server paradigm, where a client application exchanges messages with a server. In the

client-server paradigm the communication is always initiated by the client, and the server is

permanently listening incoming connections from new clients.

In order to allow clients and servers from different manufacturers to communicate, the typical

internet applications adopt a standard format for the messages, which are defined by an

application protocol. As its name implies, an application protocol is specialized for a specific

application. It is not generic as IP, TCP and UDP. Some application protocols are implemented

over UDP and other over TCP, according to the needs of the application.

The figure illustrates the application protocol called SMTP (Simple Message Transfer Protocol)

for sending email. The PDU of the application protocol is carried within the payload of the PDU

of the transport protocol according to the concept of the protocol stack.

21. TCP/IP Protocol Stack

Ethernet is not considered part of the TCP/IP protocol stack

The figure illustrates the concept of what is commonly called: TCP / IP protocol stack.

Officially, the Ethernet is not considered part of the TCP / IP stack. This is due to the fact that

the IP protocol is independent of the transmission technology. Also, TCP/IP and Ethernet are

defined by different standard organizations. TCP/IP protocols are defined by to the Internet

Engineering Task Force (IETF), while Ethernet is defined by the Institute of Electrical and

Electronics Engineers (IEEE). But conceptually, the position of the Ethernet protocol with

respect to the TCP / IP stack is that indicated in the figure.

The TCP / IP stack follows the concept of layered network model. The TCP / IP architecture is

composed of the layers: application, transport and network. Protocols that are in the same

layer perform the same function, and can be used simultaneously. By the other hand,

protocols in different layers can be used together. For example, TCP and UDP are both in the

transport layer, so only one of them can be used in a frame/packet. Moreover, as the IP is in a

different layer, it is possible to have the TCP / IP and UDP / IP combination.

As the figure shows, the position of the protocol in the stack determines how it is

encapsulated. A higher layer protocol is always encapsulated in the payload of the protocol

right below.

22. OSI Layer Model

Open Systems Interconnection Model (OSI) is a conceptual model that permits to classify

network equipments and protocols.

Many telecommunications systems follow a layered network model. In spite of being,

probably, the most famous family of protocol, TCP/IP is only one protocol stack among many

others.

In order to support a classification for the huge miscellaneous of existing protocols, a generic

reference model called the OSI model, was defined by ISO (International Organization for

Standardization). The OSI model was a work developed from 1983 to 1995. The OSI model uses

more layers than TCP / IP model. The model consists of seven layers, numbered 1-7 from the

lowest to the highest level layer.

There is no perfect match between the TCP / IP model and the OSI model. In practice, the

application layer of the TCP / IP architecture encompasses functions of the layers Application,

Presentation and Session of the OSI model. The Ethernet function is covered by two layers:

physical and data link.

The OSI model is also useful for classifying network equipment according to the protocol layer

they operate. For example, a Switch is a layer 2 device, as to forward the frames it uses the

addressing information of the data link protocol. A router, on the other hand, is a Layer 3

device as it performs its functions by analyzing the addressing information of the network

protocol.

In the market, there are devices that perform the functions of more than one layer. For

example, a Layer 3 switch is also able to perform routing.

23. Protocol Hierarchy and PDU Transportation

Each protocol transports the PDU of the protocol of the above layer. Headers are added

during transmission and removed during reception.

As shown above, the layered model determines the sequence at which protocol headers are

added and removed when sending or receiving a frame.

Generally, the unit of information produced by a layer is called PDU (Protocol Data Unit). As

discussed earlier, some PDUs receive a special designation (nickname): the Data Link PDU (DL-

PDU) is called frame, and the network PDU (NPDU) is called package. Also the Transport PDU

(TPDU) is called segment for TCP or datagram for UDP.

Ideally, the information added by a protocol of a particular layer (e.g., the network layer)

should be interpreted only by the same layer on the receptor. That is, only the piece of

software code responsible for implementing the network layer should be concerned by the

information in the network header. All other layers in the stack should be able to operate

without using this information. This is the principle of independence of the OSI model layers.

The principle of independence is not always respected by practical reasons. For example, many

routers use the information in the TCP or UDP headers to perform quality of service (giver

more priority for some packets with respect to others). It is a violation of the independence

principle.

24. Protocol Classification and Network Layer Model

Protocols can be classified according to the OSI model. Only one protocol of each layer can be

used at a time.

The OSI model allows the comparison of different protocol families. Presently, other protocol

families, such as IPX and AppleTalk, are not frequently mentioned because they have being

overshadowed by the TCP / IP protocol stack.

Unfortunately, despite having seven layers, they are not always sufficient to provide a useful

classification for some protocols. In some cases, the layered model provides more layers than

necessary. For example, the presentation and session layers are typically embedded in

application protocols.

Also, there are situations where more layers are required. For example, ATM and MPLS allow

routing packets, but are still dependent on the functions of IP routing (to update their routing

tables). For this reason, these protocols are often classified as Layer 2.5.

Technologies such as Ethernet (IEEE 802 family) and ATM have their own layered models,

which are often considered sub-layers in the OSI model.

The tunneling operation (used in VPN - Virtual Private Networks) also modify the structure

proposed by the OSI model, since protocols of the same layer may appear more than once in

the same frame, or even disrespect in the stacking order.

25. Conclusion

In this chapter we observed three main concepts:

First, LANs and WANs use different technologies for building computer networks. LANs are

basically built by computers and switches. And WANs are built by routers.

WANs can be seen as a set of interconnected LANs by a cloud of routers. This cloud of routers

that intermediate the communication of LANs is called “inter-nets” or “between nets”, what is

the origin of the famous name Internet.

Second, we have seen that the TCP / IP architecture is independent of the technology used for

transmitting frames, and introduces two levels of addressing: the IP address that allows

identifying networks, and ports numbers that allows addressing processes. The application

protocols, in turn, provides for standardization of the messages exchanged between clients

and servers in order to enable the inter-operation between products of different vendors.

Finally, we have seen the concept of network layer model, which is used to classify both:

protocols and network equipments. The layered model defines that protocols belonging to the

same layer are competitors (they are not used at the same time to transmit a frame), and

distinct layers protocols are complementary (i.e., they can be transported in the same frame).