Open Systems Interconnection ( OSI ) is a standard reference model for communication between two end users in a network.

The model is used in developing products and understanding networks. Also see the notes below the figure.

Illustration republished with permission from The manual Page .

OSI divides telecommunication into seven layers. The layers are in two groups. The upper four layers are used whenever a message passes from or to a user. The lower three layers are used when any message passes through the host computer. Messages intended for this computer pass to the upper layers. Messages destined for some other host are not passed up to the upper layers but are forwarded to another host. The seven layers are: Layer 7: The application layer ...This is the layer at which communication partners are identified, quality of service is identified, user authentication and privacy are considered, and any

constraints on data syntax are identified. (This layer is not the application itself, although some applications may perform application layer functions.) Layer 6: The presentation layer ...This is a layer, usually part of an operating system, that converts incoming and outgoing data from one presentation format to another (for example, from a text stream into a popup window with the newly arrived text). Sometimes called the syntax layer. Layer 5: The session layer ...This layer sets up, coordinates, and terminates conversations, exchanges, and dialogs between the applications at each end. It deals with session and connection coordination. Layer 4: The transport layer ...This layer manages the end-to-end control (for example, determining whether all packets have arrived) and error-checking. It ensures complete data transfer. Layer 3: The network layer ...This layer handles the routing of the data (sending it in the right direction to the right destination on outgoing transmissions and receiving incoming transmissions at the packet level). The network layer does routing and forwarding. Layer 2: The data-link layer ...This layer provides synchronization for the physical level and does bit-stuffing for strings of 1's in excess of 5. It furnishes transmission protocol knowledge and management. Layer 1: The physical layer ...This layer conveys the bit stream through the network at the electrical and mechanical level. It provides the hardware means of sending and receiving data on a carrier.

The OSI model Introduction

Network equipment manufacturers have proposed and developed network architectures specific to their devices. For instance, IBM has developed SNA, DEC has developed DNA... These architectures have all the same defect: as they are manufacturer-specific systems, it is difficult to interconnect them, unless manufacturers agree on a common architecture. Consequently, in order to avoid the development of hundreds of solution for the interconnection of these heterogeneous architectures, the ISO (International Standards Organisation), body that depends on the UNO and composed of 140 national normalisation bodies, has developed a reference model called the OSI model (Open Systems Interconnection model). This model describes the fundamental concepts and the approach used to normalize the interconnection of open systems (a network is made up of open systems when modifying, adding or removing one of these systems does not modify the global working of the network).

When designing this model, taking the heterogeneity of the equipment into account was a fundamental issue. Indeed, this model was designed to allow the interconnection of heterogeneous systems for historical and economic reasons. Besides, it should not support a particular provider. Lastly, it should make it possible to adapt to the evolution of data to process without calling into question the investments. Thus, all this led the adoption of common communication and co-operation rules between the equipment, i.e. this model should logically carry out to an international standardization of protocols. The OSI model is not a real network architecture, because it does not really specify the services and protocols each layer should use. It rather describes what the layers must do. Nevertheless, the ISO has developed its own standards for each layer, and this independently of the OSI model, i.e. as does any manufacturer. The first works related to the OSI model date from 1977. They were based on the experience gained in the area of wide area networks and local private networks; the OSI model was indeed supposed to be valid for any type of network. In 1978, the ISO proposed this model as the standard ISO IS7498. In 1984, 12 European manufacturers, joined in 1985 by the main American manufacturers, adopted this standard.

The different layers of the OSI model

The 7 layers
The OSI model is composed of 7 layers:

The principles that led to these 7 layers were the following:

A layer must be created every time a new level of abstraction is necessary, every layer has well defined functions, the functions of each layer must be chosen in the objective of the international standardization of protocols, boundaries between layers must be chosen so as to minimize the flows of data through interfaces, the number of layers must be such as there is no cohabitation of completely different functions within the same layer and such as it is not too difficult to control the architecture.

The low layers (1, 2, 3 and 4) are necessary to the routing of information between the two concerned ends and depend on the physical medium. The higher layers (5, 6 and 7) are responsible for the data processing relative to the management of exchanges between information processing systems. In addition, layers 1 to 3 intervene between close machines, but not between ending machines that can be separated by several routers. On the contrary, layers 4 to 7 intervene only between distant hosts.

The physical layer

This layer is in charge of the raw transmission of bits over a communication channel. This layer must guarantee the perfect transmission of the data (a bit set to 1 must be received as a bit set to 1). Concretely, this layer must standardize the electrical characteristics (for instance, a bit set to 1 is represented by a voltage of 5V), the mechanical characteristics (the shape of the connectors, topology...), the functional characteristics of the circuits of data and the procedures of establishment, maintenance and release of the circuit of data. The typical information unit for this layer is the bit, represented by a given voltage.

The data link layer

Its has a role of "binder": it will transform the physical layer into a connection a priori freefrom transmission errors for the network layer. It splits the input data of the sender into frames, sends these frames in sequence and manages the acknowledgement frames sent back by the receiver. To remind, for the physical layer, the data do not have any particular meaning. The data link layer must therefore be able to recognize the limits of frames. This can actually pose problems, since the sequences of bits used to identify boundaries may also appear in the data. The data link layer must be able to signal a transmission problem by sending an appropriate frame. In a general way, an important role of this layer consists in detecting and correcting errors that occured on the physical layer. This layer integrates also a flow control function to avoid the blocking of the receiver. The information unit for this layer is the frame made up of a few hundreds to a few thousands of bytes maximum.

The network layer

This layer is in charge of the sub-network, i.e. the routing packets over the sub-networks and the interconnection of the various sub-networks. When designing it, it is very important to determine the routing mechanism and calculation of the routing tables (static or dynamic tables...). The network layer also controls sub-network congestions. It is also possible to complete it with accounting functions for invoicing on volume, but this may be delicate. The information unit for this layer is the packet.

The transport layer

This layer is responsible for the good delivery of messages to the recipient. Its main role is to take the messages of the session layer, split them into smaller units and give them to the network layer, while checking pieces arrive correctly. Therefore, this layer also re-assembles the initial message when it receives the pieces. This layer is also responsible for the optimization of the network resources: normally, the transport layer should create a network connection for every transport connection required by the

session layer, but it is able to create several network connections by session layer's process, for example to improve the bit rate. Conversely, this layer can use one network connection to transport several messages at the same time, using multiplexing. In any case, all this must transparent for the session layer. This layer is also responsible for the type of service to provide to the session layer, any finaly to the users of the network: connection-oriented or connectionless service, with or without guarantee of the delivery order, broadcast... Thus, this layer is also responsible for opening and closing network connections. One of its latest role is flow control. It is one of the most important layers, because it provides the basic service to the user and controls the whole connection process, with all the related constraints. The information unit for this layer is the message.

The session layer

This layer sets up and synchronizes the exchanges between distant processes. It binds logical addresses to physical addresses for distributed tasks. It also binds two application programs that must cooperate control their dialogue (which one should speak, which is currently speaking...). In this latter case, the service of set up is called the token management. The session layer also makes it possible to insert recovery points in the data flow in order to resume dialogue after a failure.

The presentation layer

This layer deals with the syntax and semantics of the transmitted data: it processes the data so as to make it compatible between communicating tasks. It will ensure the independance between the user and the data transport. Typically, this layer can convert, format, crypt and compress the data.

The application layer

This layer is the point of contact between the user and the network. Therefore, it brings the basic network services to the user, such as file transfer, electronic mail...

The data transmission through the OSI model

The transmitting process gives the data to be sent to the receiving process to the application layer which adds to them an application header AH (AH may be empty). The result is then transmitted to the presentation layer.

The presentation layer then transforms this message and adds (or not) a new header (possibly empty). The presentation layer does not know and does not have to know the possible existence of AH; actually, for the presentation layer, AH is part of the user data. Once the data processing is finished, the presentation layer sends the new "message" to the session layer and the same process starts again. The data then reach the physical layer which will indeed transmit the data to the recipient. Once received, the message will go up the layers and the headers are gradually removed until it reaches the receiving process:

The most important concept is as follows: it should be considered that each layer is programmed as if it were really horizontal, i.e. as if it dialogued directly with its receiving peer layer. When dialoguing with its peer layer, each layer adds a header and sends it (virtually, thanks to the subjacent layer) to its peer layer.

Networks classification Technique of transmission

A first manner to classify networks is to consider the technique of transmission:

broadcast, point-to-point (peer-to-peer).

Broadcast networks are characterized the following way: all machines share the same transmission (logical) channel. Therefore when a machine sends a message on such a network, all machines without exception receive it. This is why we speak about broadcast. Machines on such a network are identified using what we call addresses. A variant to this kind of network is multicast network: only designated machines can receive the sent message. To get the message, machines must subscribe to a multicast group. On the other hand, point-to-point (peer-to-peer) networks are characterized by communication channels that link only 2 specific machine, that is to say that a message must jump from one machine to another to reach its destination. Thus, in a general way, we can consider that networks with limited dimensions (among others local networks, see below) use preferably broadcast, though wide networks preferably use a point-to-point structure.

Classifying using networks dimensions

General classification
We'd rather classify networks and multiprocessing systems according to the interconnection distance of processors. We then establish the following classification: distance < 1cm < 1m < 200m < 1km < 10km category massively parallel machines or data flow machines multiprocessor machines Local Area Networks (LAN) campus networks Metropolitan Area Networks (MAN)

< 1.000km Wide Area Networks (WAN) > 1.000km interconnection of wide networks (Internet)

This classification highlights also a technical classification. Technical solutions involved to make use of them are indeed quite distinct to each other. For instance, connections on a local network are realized with coaxial cables or twisted pair cables, though connections on a WAN are realized with optical fibres.

Local Area Networks (LAN)

The main characteristics of such networks are:

restricted geographical expanse (less than 200m), high bit rate (commonly between 10 and 100 Mbps, but it can be greater), global integration: it is managed by a unique organization.

This kind of networks is usually used to connect computers of a company. Such networks depend on a private authority. Because of the small size of this kind of networks, transmission periods are short, with few errors. All this make it easy to administrate.

Metropolitan Area Networks (MAN)

We call them "metropolitan networks" because they are usually used on areas like cities. Main characteristics:

middle-size geographical expanse (diameter shorter than 10km), high bit rate (between 10 and 100 Mbps), interconnection capabilities with local networks.

This time, such networks depends on a public authority. We can use this kind of networks to transmit voice and data. These networks are usually built with 1 or 2 transmission cables (bus) without routing devices. It is a very simple structure, as all computers are indeed directly connected to the bus:

DQDB architecture

We then use the DQDB standard (Distributed Queue Dual Bus) for transmissions.

Wide Area Networks (WAN)

This type of networks covers a wide geographical area, sometimes as large as 1.000 km. Actually, there is no dedicated infrastructures: we directly use those for telecommunications. It is therefore a meeting point for computer science and telecoms. These infrastructure are currently being renewed to complete this convergence and improve the quality of services provided on such networks. The problem is that it is a very expensive operation (see below, performance problems). In a more technical point of view, these wide area networks are developed around switching nodes (routers or switches) connected to each other by lines. These nodes form what we call switching subnetworks or simply subnetworks. The main goal of these subnetworks is to interconnect hosts (computers). These hosts are usually connected to a LAN, which is connected to a subnetwork via a router:

WAN: subnetwork/host relation Routers are often connected to each other with other routers. In this case, a packet received by a router is forwarded to the next router only when it has been xompletely received and the first router is able to reach the second one. This working characterize what we call a point-to-point, packet switching or store-and-forward subnetwork. Apart from satellite networks, all WAN are of this type. As they are wide, these networks are subjects to some reliability (the background noise grows with the length of lines) and performance (transmission times also grow with the length of lines) constraints and problems. This generate a non negligible increase of costs when we want to improve the quality. Fortunately, the appearance of efficient and cheap synthetic optic fibres can help reducing installation costs. Examples of WAN: Arpanet (the very first wide network that initiated the Internet), Internet, Transpac...

Other classification
The improvement of technologies, and therefore bit rates, suggests a new classification based on their bit rate (a more "logical" classification). We then have 3 categories:

low and medium bit rate networks (< 200 kbps), high bit rate networks (200 kbps < bit rate < 20 Mbps), very high bit rate networks (> 20 Mbps).

Local networks are getting faster and faster, especially with the coming of Fast Ethernet, that has become cheap and very reliable.