OSI Reference Model

The Open Systems Interconnect Reference Model is a seven layer model defined by the ISO (International Standards Organization) to create an interoperable network. The layers describe all functions of a network from the physical medium up to the application.The idea of the ISO was to create separate layers independant of each other. This way it would be possible to use the best or most suitable layer from every protocol. This didn't succeed, but the OSI Reference Model is still the best way to describe a network.
OSI Reference Model
7 Application Layer
6 Presentation Layer
5 Session Layer
4 Transport Layer
3 Network Layer
2 Data Link Layer
1 Physical Layer7 Application Layer

The application layer contains all the management information for the distributed applications.The data units here is the real user data. Active devices for internetworking are called gateways.Note: This is not according to OSI which calls a gateway a device which is active on layer 3

6 Presentation Layer
The presentation layer is responsible for formating data in such way that it is ready for presentation to the application. This means translation of different character formats (ASCII/EBCDIC) is done here, but also text (de)compression, virtual terminal emulation and encryption/decryption. It is completely responsible for translation, formatting and the syntax selection.The data units on this layer are called messages

5 Session Layer
The session layer is responsible for establishing, maintaining and termination of session between applications. The session layer establishes and sets and synchonizes the parameters for dialogues between two devices.

4 Transport Layer
The transport layer is responsible for a guaranteed delivery of data. It provides a reliable service, it takes care of the setup, maintenance and shutdown of vitrual circuits and it is responsible for fault detection, error recovery and flow control.Data units on this layer are refered to as datagrams

3 Network Layer
The network layer is responsable for connectivity and path sellection between two end systems. By using the network layer it is possible to separate LANs into distinct networks. There is no guarantee of correct data delivery, since the network layer does not provide error correction.The devices active on this layer are called routers and the data units are refered to as packets.Note: There is a slight distinction with official OSI name for active devices on this layer. The OSI talks of gateways, but the OSI is probably the only one, we will refer to routers.

2 Data Link Layer
The data link layer is responsable for the physical addressing of the data. This means it's addressing is based on the true network interface card addresses, also known as MAC-addresses. The data link layer is also responsable for error notification, ordered delivery of frames and flow control.The IEEE has split this layer up into MAC and LLC, and even added SNAP.

The reason for this splitup is that the LLC, with or without SNAP, is identical for all networks and the MAC-layer is network specific. Ethernet and Tokenring have a different medium access control protocol (CSMA/CD and Tokenpassing) but the same LLC (IEEE 802.2)It's on the MAC part of the data link layer where bridges are active. The data units related to this layer are called frames.

1 Physical Layer
The physical layer is responsable for activating, maintaining and deactivating the physical link between to end stations. It defines electrical (voltage levels, timing and coding of voltage levels), mechanical (physical conntectors and cabling) and functional (data rates, max. transmission distances) specifications.Devices active on this layer are called repeaters. The units transmitted are bits (1 or 0).

Circuit Switching Vs Packet Switching

Circuit Switched Dial-Up Lines
Circuit switched, or digital dial-up (switched 56, ISDN) links form on demand. For the duration of a dial-up session, they behave as open pipes. Within the region of a service, the cloud that defines the boundary of that service in principle includes all end points. Establishing a link requires only a few seconds. Redundant paths exist among end points. Peer interaction among end points requires that the number of links grows beyond the number of end points.

Packet Switched Lines
Links from end points to the 'cloud' (boundary of the packet-switching service) (X.25, frame relay) are usually open pipes.Within the cloud, connections among end points are usually PVCs. Configuring a new PVC, which requires no equipment changes, can take as little as a few hours.Redundant paths can exist among end points. For every-to-every connectivity, the number of links into the cloud does exceed the number of end points (however, the number of PVCs must grow).

ARCnet

Introduction

To start off, ARCnet is not a standardized network. All information provided here is gathered through the years by different people and by just making mistakes, so if you have other experiences don't be surprised, but share them with us. This way we can all learn from it.
ARCnet (Attached Resource Computer NETwork) was originally developed by Datapoint Corporation in 1968, way before OSI and the like. The original purpose was a harddisk interface, later it became popular as a local area network.
According to the Datapoints ARCNET Designer's Handbook 61610 © 1988, ARCnet is: a high-performance, bus-structured, token-passing, base-band network linking two or more computers through high-speed connections. The network can support up to 255 nodes, transmitting data at a rate of 2.5 Mbps up to 4 miles in distance (when active hubs are used between nodes).
ARCnet uses a Token Passing bus structure much like IEEE802.4. It's most common speed is still 2.5 Mbps, but we know of versions runnig 20 Mbps (ARCnet plus by SMC, NCR and Datapoint) and the rumour is there is even a 100 Mbps (by Thomas Conrad) version. So it seems like it is still in progress. The reason for this is probably its easyness (ARCnet is very tolerable) in cabling and the predictable delivery of packets through the use of a token architecture.

The ARCnet devices

ARCnet knows 5 different types of devices:
Active hub
They split and amplify the signal. The are a bit like the hubs in Ethernet. Unused ports on an Active Hub do not need termination, although it is advised to terminate them. The active hubs provide electrical isolation at each port. They feature problem detection and segment partitioning, so only the segment on that port will be affected by problems of one of the attached devices to that segment.
Active Link
Which actually is a repeater. It has two ports and acts as a two port active hub.
Passive hub
They just split the signal, which means if it is a four port hub, every port gets 1/3 of the signal (one port was incoming). Unused ports on an Passive Hub do need termination. A Passive Hub is a small box with 4 BNC connectors and an internal resistor network.
Bus NIC (RIM)
A network card with a high-impedance driver (SMC 9068), only suitable for bus-segments.
Star NIC (RIM)
A network card with a low-impedance driver (SMC 9058), only suitable for a star-segments.

Local Loop

IntroductionThe Local Loop is the line from the customer premisis to the local telco office. This line connects you to the nearest facility of the local PTT. The cable used is generally 26 AWG unshielded twisted pair cable (but it can be fiber also), with runs of about 2.5 miles or 4 km length. Most lines do carry only a single call, but more calles are possible (ISDN, xDSL). The Local Loop is terminated at two points. One is the connection at the customer premisis and the other is the connection to the channel bank or mux at the local telco office.

There are two types of connections:

Analog
Digital

Analog lines
Analog lines are always copper. The lines could be used for voice and data and can be dial-up or leased lines. The specificness about these lines is that the information is transported through the wires in an analog way, meaning sine waves are used and not ones and zeros, like in digital lines. The terminals on these lines are often telephone sets or modems. For a description of these devices see their corresponding pages.

Loading Coils
To assure electrical consistency over the entire analog copper line loading coils are placed in the line. Loading coils can not be used in digital links, because they absorb digital pulses.

Digital lines
Digital lines may be copper or fiber.

DTE and DCE

DTE and DCE
The terms DTE and DCE are very common in the datacommunications market. DTE is short for Data Terminal Equipment and DCE stands for Data Communications Equipment. But what do they really mean? As the full DTE name indicates this is a piece of device that ends a communication line, whereas the DCE provides a path for communication.
Let's say we have a computer on which wants to communicate with the Internet through a modem and a dial-up connection. To get to the Internet you tell your modem to dial the number of your provider. After your modems has dialed the number, the modem of the provider will answer your call and your will hear a lot of noise. Then it becomes quiet and you see your login prompt or your dialing program tells you the connection is established.Now you have a connection with the server from your provider and you can wander the Internet.
In this example you PC is a Data Terminal (DTE). The two modems (yours and that one of your provider) are DCEs, they make the communication between you and your provider possible. But now we have to look at the server of your provider. Is that a DTE or DCE?The answer is a DTE. It ends the communication line between you and the server. Although it gives you the possibility to surf around the glode. The reason why it is a DTE is that when you want to go from your provides server to another place it uses another interface. So DTE and DCE are interface dependend. It is e.g. possible that for your connection to the server, the server is a DTE, but that that same server is a DCE for the equipment that it is attached to on the rest of the Net.

Multiplexing

Multiplexing
GeneralA multiplexor makes it possible for several devices to share a single communication line. Every device has a point to point connection to a device on the remote multiplexor, while there is just a single link available.The way the main link is shared is three folded:
Frequency Division
Statistical
Time Division
Multiplexors do not have any intelligence about the data they are multiplexing. They are interface dependend. The input ports are usually called channels and the output port is called the composite or main port.

Frequency Division Multiplexing:
FDMs devide the available bandwidth (Hz) of a link into multiple sub-channels with a smaller bandwidth. A good example is the way cable TV (CATV) is broadcasted to every home. A single cable contains all the channels you can choose from on your TV-set.Guardbands are used to separate the sub-channels. This means there is some overhead. This kind of multiplexing isn't used often (meaning in a wide range of applications), because FDMs can be as easilly expanded as the other multiplexing technics.

Time Division Multiplexing:
TDMs use the full bandwidth for every channel, but not at the same time. In a round robin fashion every channel gets its time slice (time slot) to the shared link.Each channel is sampled for a certain amount of time (the sample time depends on the number of channels and the input speed, length of a bit). The state of the channel is then send to the remote TDM where it is demultiplexed from the incoming bitstream and send to the corresponding channel. Because the sampling rate is a multiple of the bit time every bit is sampled more than once to prevent data loss. The aggregate rate of the channels can not exceed the rate of the composite port.When channels are not used, their bandwidth is still reserved (an empty slot is send). A more efficient way of multiplexing is statistical multiplexing.

Statistical Multiplexing:
A STM tries to use the capacity of the line as optimal as possible. Every channel is buffered and only those channels that have something to send are multiplexed and send to the remote side. This requires some intelligence from the muxes and a way of indicating which data came from which port.A large amount of data or many 'used' ports will soon flood the mux or the shared link.

X.25

Overview
In 1976 the CCITT adopted the X.25 standard labeled "Interface Between Data Terminal Equipment (DTE) and Data Circuit Terminating Equipment (DCE) for Terminals Operating in the Packet Mode on Public Data networks". X.25 is a peer-to-peer network and is only an interface protocol to packet-switched public networks. Packet-switching techniques are a flexible alternative that provides statistical multiplexing (the ability to have more than one logical channel over one physical channel) and any-to-any connectivity. The DTE interfaces with an X.25 service provider (DCE). Packet switches within the X.25 network are refered to as Data Switching Equipment (DSE).
X.25 specifies standards for the bottom three layers of the OSI-model.
1- The physical layer requires the use of interface recommendations X.20 (asynchronous) and X.21 (synchronous).
2-The link layer is split in two separate procedures based on ISO 3309 High Level Data Link Control (HDLC):
-Link Acces Procedure (LAP)
This is an original X.25 provision consisting of the HDLC "asynchronous response mode".
-Link Acces Procedure Balanced (LAPB)
This one was added later and is similar to the HDLC "asynchronous balanced mode". The LAPB makes it possible for communicating parties to operate in an autonomous, balanced mode in which neither is considered a master or slave.
-The network layer is called the packet level in X.25. This packet level accomplishes the flow of control information and data across the network.


Layer2 Layer 3 Layer 2
Flag Address Control Information Frame Check Sequence Flag


The frame format of X.25 is according to the ISO 3309 standard. The FLAG field is used to identify the beginning and ending of a frame. The ADDRESS field designates the binary address of the receiving station. The CONTROL field indicates the type of frame, such as:
-Unnumbered frames, consist of command information to start or stop the communication link. They do not contain data.
-Supervisory frames, contain routine operational commands, including the response to previously received frames. They do not contain data.
-Information frames, contain data packets, and may also contain response information

The INFORMATION fields contain the actual data and a data header, or is empty
The data header is 24 bits and split up in 4 parts:
-4 bits General Format Identifier (GFI)
-4 bits Logical Channel Group Number (LGN)
-8 bits Logical Channel Number (LCN)
-8 bits Packet Type ID (PTI) or P(R)MP(S)0 (Packet sequence numbers used when a packet contains user data)

The FRAME CHECK SEQUENCE is the CRC (Cyclic Redundancy Check) code.

The Transmission
The transmission is started in a virtual circuit operation with a call request. The initiating DTE asks the network for a virtual circuit with the destination DTE. The packet used for this is called a Call Request packet. The Call Request packet contains the destination address.The address (along with the originating DTE's identification) is used by the network for the duration of the call.The network responds with a "connect" message or with a message saying that the circuit can't be established and usually why. Now the actual data transfer can begin, on a routine or interrupt basis, and according to a set of "flow control" constraints imposed by the network. All these activities are handled by specific packet types. X.25 defines more than 30 packet types for various functions.

The Layer Discriptions
The Physical Layer:This one is not specified in the CCITT recommendation, but X.25 is traditionally used on leased lines, although there is a dial-up function which uses X.121 as an addressing scheme. This addressing scheme allows up to 14 digits arranged in a hierarchy of zones and domains.The connection is synchronous with speeds of 2.4, 4.8, 9.6, 56 and 64 kbps. There is also a 2 Mbps version.Common RS-232/EIA interfaces can be used, but more usually you'll find X.21. For asynchronous connections you'll need a Packet Assembler/Disassembler (PAD).

The Data Link Layer:
The data link layer is responsible for the packet format, which is described above. And will be discussed here in more detail.
-The first and last field of the packet is the FLAG (X'7E'). -The ADDRESS field has two possible values X'01' and X'03' which are also known as A and B. Which are used to indicate 'originate' and 'answer'.
-The CONTROL field is initialized with a Set Asynchronous Balanced Mode (SABM). There is also an extended operation (SABME) that provides a modulo-128 frame window. LAPB is a full-duplex mode of operation.

The Network Layer:
Often referred to as Packet Layer Protocol (PLP). The layer protocol provides a statistical multiplexing technique through the use of multiple logical connections across a physical link.The three possible logical (virtual) connections are:
-Logical Channel
-Switched Virtual Circuit (SVC)
-Permanent Virtual Circuit (PVC)

The PVCs and SVCs are both end-to-end connections between two users, but they are different in the fact that PVCs are predefined by the X.25 service provider and SVCs are not. PVCs also have the advantage that they do not require the connection setup and takedown time which are needed with SVCs.

The network layer is also responsible for the information field, which consists of two parts: the actual data and the data header.The first part of the data header is the 4-bit General Format Identifier (GFI) which indicates (among other things) the presence or absence of data.The 4-bit Logical Channel Group Number (LGN) is the group number.The 8-bit Logical Channel Number (LCN) is the channel number.The LGN and LCN combined can be used as one big channel number.LCN 0 is normally used for emergency network commands.And the last part is the 8-bit Packet Type Identifier (PTI) or Level 3 flow control sequence numbers. The PTI is only present at level 3 protocol packets.

ATM (Asynchronous Transfer Mode)

Asynchronous Transfer Mode works with very short, fixed-length cells @ 44.7Mbps to 2,4Gbps and more. That allows for time-efficient and cost-effective hardware (switches, etc.). ATM uses 53 byte cells, consisting of a 5 byte header and a 48 byte payload. Because ATM is connection-oriented the cells can have such a short adress space and the cells are not used for establishing the circuit and maintaining it. Once a circuit is set up the bandwidth can be used entirely for data transport. After the circuit is set up, ATM associates each cell with the virtual connection between origin and destination. This can be a virtual channel or path. The 40 bit header holds 8 bits for the virtual path (256 max), and 16 bits for the virtual channel (65536 max). Having both virtual paths and channels make it easy for the switch to handle many connections with the same origin and destination.
The proces that segments a longer entity of data into 53 byte cells is called 'segmentation and reassembly' (SAR). The data that goes into these cells comes from different native mode protocols, such as TCP/IP. The ATM Adaptation Layer (AAL) takes care of the differences between the different sources. The AAL adapts the protocols to an ATM intermediate format. It uses socalled 'classes' to do so. AAL type 3 and 4 handle transmissions of connectionless data, AAL type 5 is intended for connection-oriented services.
ATM relies on different classes of service to accomodate different applications (voice, video, data). They define the bits and bytes that are actually transmitted, as well as the required bandwidth, allowable error rates, and so forth. Class A and B, have timing compensation, for applications that cannot tolerate variable delays. Class C and D, no timing compensation, for data applications like LAN interconnect. Class D also simulates connectionless communicaations, comonly found on LANs.

FDDI and CDDI

General

In 1989 the ANSI adopted a standard X3T9.5 which became known as FDDI (Fiber Distributed Data Interface), a 100 Mbps network over fiber. FDDI describes the layers one and two of the OSI model with the connectors used and the characteristics of the fiber cable. In 1994 there was also a copper part adopted into the standard called CDDI (Copper Distributed Data Interface).FDDI and CDDI both are ring topologies using a token for media access, although star connections through concentrators are possible (compare Token Ring and the use of MAUs).
X3T9.5 defines a token passing ring, dual fiber network. It is designed to be a fault tolerant network. What is called THE ring is actualy a dual ring of which one is used and the other is the 'spare' one.All connections to the network are made through MIC (Media Interface Connector) connectors for FDDI multi-mode fiber and through ST connectors for single mode fiber. RJ45 connectors are used for CDDI.



X3T9.5 defines two types of devices:
-Stations
-Concentrators

And three ways of connecting devices:
-Dual Attached Stations (DAS)
-Single Attached Stations (SAS)
-Dual homing



Stations

Stations are devices that play a real active role, like computers, printers, etc. Stations can de Dual Attached and Single Attached.

Concentrators are devices that make it possible to connect SAS to the network. When a SAS is powered down the concentrator ensures the continuety of the ring by bypassing the port the station is attached to. They function like a hub with a star wired topology. Concentrators are, when connected to the ring, always Dual Attached, DAC (Dual Attached Concentrator).A DAC has three different types of ports A, B and M. Port A connects the input of the primary ring and the output of the secondary ring and port B connects the output of the primary ring and the input of the secondary ring. The M port connects to the S (slave) port of a SAS.A SAC (Single Attached Concentrator) has only two different ports: S and M. The M ports connect to SASs and the S port connects to a DAC or another SAC M port.

Dual Attached Stations

(DAS)Stations connected to the ring are always Dual Attached and need to be up and running all the time. DAS have two ports, A and B. Port A connects the input of the primary ring and the output of the secondary ring and port B connects the output of the primary ring and the input of the secondary ring.DAS can also be connected to concentrators. This way the connection is fault tolerant, but the station need not to be up and running all the time.

Single Attached Stations

(SAS)These devices may not be connected to the dual ring. They only have one port and should always be connected through a concentrator. A SAS has a S-port (Slave) which is connected to the M-port (Master) on the concentrator.

Dual Homing

This means a concentrator or a DAS can be connected to two concentrators. One link will be active, but if that link fails the other one will take over. So actually you create two SAS links. But instead of adding two SAS cards you use a DAS card.

The SMT part is a generalThis document is build to describe FDDI. For the CDDI version see the corresponing document. The LLC and MAC descriptions are identical so the only differences will be in the TP-PMD and TP-PHY.

LLC; Logical Link Control

The LLC is identical to IEEE 802.2.

MAC; Medium Access

ControlFDDI uses a Token-passing procedure to gain access to the physical medium. Much like Token Ring a station is only allowed to transmit data when it possess the token.A token is a special frame that runs around the network and a station that has something to send and receives the token removes the token from the network and places it's data instead on the network. The station for which the data is intended copies the data into it's buffers. When the data is back to the sender, that one removes the packet from the network.A station may only hold the token for a certain amount of time, which is called the Token Holding Time (THT). After that it must stop transmitting data and must release the token just after the last packet send.
When a FDDI network is initialized every station measures the Token Rotation Time (TRT). The MAC protocol limits the maximum token rotation time to a value called the Target Token Rotation Time (TTRT). From this every station calculates the Token Holding Time by the formula:THT=TTRT-TRT
FDDI also has the possibility to use priorities. First it distinguises data as being Synchronous or Asynchrounous.Synchronous data is data that is time sensitive like voice and video. When a station captures the token it first sends the the synchronous data.Asynchronous data can be devided into eight priority catagories. And will be send in importance order after the synchronous data has been send.

PHY

The PHY is responisble for coding and timing of the signalsThe actual speed used on a FDDI ring is 125 Mbps, since one control bit is added for every four data bits. That means that the frequency used on the fiber cable is 62.5 MHz, because there is one 1->0 transition per Hz.

PMD; Physical Medium Dependent

InterfaceThe PMD is responsible for the conversion of electrons to lightThe first version of the PMD defined multi-mode fiber optic cable with a maximum of 2 km. between two stations. Later a single mode version was added increasing the distance to 40 km.

SMT; Station Management Task

This function comes into action during network initialisation, when generation the token. The SMT monitors the PMD, PHY and MAC functions and is responisble for:
RMT: Ring Management
CFM: Configuration Management
CMT: Connection Management
PCM: Physical Connection Management
ECM: Entity Coordination Management
One of it's functions is also to check the optical bypass for switched off stations. It also generates status reports about the network and it's attached stations.
The SMT is a seperate protocol which can be reached through the MAC address of the station and thus the station can me monitored or remotely modified (only the station characteristics according to the network).

CDDI
CDDI is the copper version of FDDI. Since the differences are only in the PMD and the PHY, this chapter only describes the TP-PMD and TP-PHY.
PHYThe PHY is responisble for coding and timing of the signalsThe coding used is a MLT-3 line coding, which is a three-voltage-level encoding scheme.The transmission of a binary 0 is represented by sending the same voltage that was in the previous bity slot.The transmission of a binary 1 is represented by changing the voltage in a controlled way from that send in the previous bit slot.The net effect is that 8% of the signals power is shifted to frequencies below 30MHz, which results in lower attenuation and less cross-talk.
TP-PMD; Twisted Pair PMDThe PMD is responsible for the actual transmission on the cable.It defines the use of CAT5 cable with a maximum length of 100 m.

Layer 3 Switching

The reason

The idea of switching on layer 3 was first proposed by a company called Ipsilon. Different standards emerged and at the moment of writing this document there is still no definitive standard. There are some solutions though that are incorporated in a standard (like ATM). For detailed descriptions see the vendors Web-Sites mentioned.
The problems, or network history in an eye-wink

Before looking at the solutions it is good to see what is causing the problems.First, when networks emerged there was a need to extend the maximum distances, so repeaters were invented. After that, people wanted to connect LANs over Wide Area Networks (WANs). This problem was solved by building bridges. When more and more networks were linked together routers were needed to make the whole thing managable.And when we had that we wanted more speed. More speed meant creating switches. The first were just called switches and switched on OSI layer 2, so they were actually switching bridges and now we need more speed and have real time traffic and need 3rd layer switches, which are actually switched routers.But since a router is situated on layer 3 it adds delay and there is another problem with routers, they route. This means that data can follow different routes to their destination with as a result that some packages will arrive earlier than packedges send before them. This makes it impossible to run time sensitive applications.
The solutions

The problems that need to be solved is the different routes for packages from the same job and the delay added by the routers.Solving different routes seems to be the easiest one. Just take a fixed path and the problem is solved, but implementing this is the real problem.Solving the delay factor, once there is a fixed path, is actually a piece of cake, because with a fixed path bridging is all that is needed.
The propositions for call setupThe solution for a fixed path is a call setup, where there is a detection of the destination and an agreement on the path that is used for the time the link is used. The technics used sound like ATM and that is were everyone agrees on; ATM will be the future. But in the meantime we have an installed base of several thousands of routers which are connected through dedicated lines and no ATM. So there is a need for a temporary different solution.
Let's see what the posibilities are on the 3rd and 2nd layer of the OSI reference model:

Layer 3 Route everywhere
Route once, switch afterwards
Layer 2/3 Switch here you can, route where you must
Layer 2 Switch everywhere

The first option, route everywhere, is the actual router as we know it today. And the last one, switch everywhere, is a real bridge. So those two options are not our concern in this document.
Route once, switch afterwards VS Switch where you can, route where you must

Those two options have a basic different view of the network. The switch where you can device sees the network as being flat and is mainly a bridge. Only when it isn't able to determine where the data must go it will use it routing capabilities.The otherone is mainly concerned about routing. It accepts the network as a must route network. Only when it has the knowledge of the path the data ought to go it will switch.Later on we will see the importance of those two different views of the network.
The technologies

The now known architectures can basically be split up into three different functions:
-Routing switch
-Flow switch
-Switched router

The Conventional Router This one is added to give you a good comparison. The route calculation and packet processing take place in the software on layer 3. This means that packets need to be moved from the layer 2 hardware interface to layer three.The information exchange between the two routers are the standard routing protocols

The Routing Switch Routing calculations takes place at layer 3 in hardware or software, while the actual packet processing takes place at layer 2. They are mostly just like ordinary routers. The speed gain is accomplished by reducing the amount of features supported and moving as much logic as possible into hardware.The information exchange between the two router switches are the standard routing protocols.

The Flow switch Route calculation and packet processing takes place at layer 3 until a flow is detected. The flow is then switched at layer 2 through the network. This way of handling data traffic much looks like ATM or frame relay with a VC for long lived traffic.The information exchange between two flow switches is accomplished through flow management protocols.

The Switched router Route calculation and packet processing takes place at layer 3. By adding a tag to the packed information the amount of layer 3 processing is reduced. According to the Cisco view you need tag-edge routers that add the tag to non-taged packets and in the rest of the network you need tag-switches which actually give you the speed gain.The information exchange between two switched routers is accomplished through the tag distribution protocol.

Repeaters

Repeaters
A repeater is a device that takes the signal off the cable at one interface, retimes it and sends it out on the other. Inside a repeater are two or more tranceivers and some logic that connects the tranceivers together. A repeaters function is to forward each incoming signal, without modification or unnecessary delay, to all ouput ports. This means it also send out collisions, fragments, jam signals, and bursts of noise.Repeaters can have two or more ports. A good example of a multiport repeater is a HUB in ethernet networks.
Repeater ports do not have to be identical. You could repeat a copper signal to a fiber signal, but that is only possible when it is the same topology. So you could repeat 10Base-T signals to 10Base-F, but it is impossible to build an ethernet to token ring repeater, such a device is called a bridge.
Since repeaters only consist of tranceivers and they do not have any knowledge about the signals they are repeating, this device is placed at OSI layer 1.
Note on Repeaters:A special kind of repeaters are called Fan-Out unit or Delni (Digital Equipment Corporation). These devices are repaters but without the PHY (tranceiver) so you only have AUI connectors on the in and output ports.

Routers

Introduction
According to the OSI-model the router doesn't exist, but is called a gateway. Since the meaning of a gateway in everyday speach is different, we will use the term router to address a device that corresponds to the following features:A router connects networks, thus working on the network layer of the OSI-model (Layer 3). It is protocol depended and the network address on one interface is different from that on another interface.

The most common application for routers is to connect LANs through WAN-links.
These routers can act in two different ways:
-Routers connect to eachother by using the WAN-link as a separate network. This means that to the router the WAN-link is an entire different network. This way you need at least three ranges of network addresses. One range for LAN1, one for LAN2 and one for the WAN-link.
-The other option is that the routers use a propriatery protocol over the WAN-link. The most common form is to use PPP. In this last example we say that the two routers and the WAN-link form one router. And both routers are called half-routers.
Usage

Routers are used to connect LANs or coupled LANs. Their means is to provide a controlled way for the delivery of packets over WAN-links and to separate LANs and/or protocols. Since a router functions on layer 3 of the OSI model, it depends on routable protocols (not all protocols are routable). The protocols need to have a way of having network numbers on which a router can base its descision to route a packet.
Another advantage of routers is that different paths can be provided to one target and the most efficient (read fastest) path can be choosen to deliver a message. To do this, routers needs to know what the network around them look like.
Through this path selection method a router can provide redundant paths to one location. This way a link-failure will not stop the connectivity. A router will make sure the damaged link is not used anymore and a different path is chosen to send the packets.
How a router works

Since routers are working on layer 3 of the OSI model they are capable of analyzing a packet up til the network address-envelope. Let's assume that a packet looks like this:
_______________________________

MAC-address destination-address data

_______________________________

This is just a very simple example and no real life packet.
The router strips the MAC-address and reads the destination-address. It then examens this address and decides if it is on a network on one of its local ports. If so it attaches the MAC-address of the station to the packet and sends it to the port where the station is on.If it is not a station on one of its local attached networks it will look at the network part of the destination-address. And chooses the port to send it out on. The packet will be send to another router and the MAC-address of that router will be attached to the packet. The next router does the same and eventually the packet will reach its destination.
In basic this is what routers do. In everyday life routers need to know what port should be used for which network and what the alternative routes are to reach that network and which stations are on which port and so on. To know all this routers need to have a way to communicate to eachother. And since there are a lot of different vendors, there need to be a standard to do this so every router can be connected to any other router.
To make a first distinction: you can split routers into two different kinds. One type uses routing tables and one type does not. To look at the non-routing-table routers first: You have three different types:
Flooding Routers
The received packet will be send to every port, so the packet will always reach its destination in the shortest possible way. All later incoming packets will be trashed by the end-station.
Random Routers
The incoming packet is transmitted to a port. Just at random and you hope the packet will arrive sometime.
Hot Potatoe
The incoming packet will be send to the port with the shortest queue. This means the router needs less buffers, but again it is not certain the packet will ever arrive.
All these techniques are old, and are rarely seen anymore. Most modern routers use routing-tables to make sure the packets will arrive save and as fast as possible. These routing-table routers are there in two flavours and most of the time the two techniques are comined:
-Static routers
-Dynamic routers

Static routers have their routes keyed in by a system administrator, this can be a lot of work if the network is large.

Dynamic routers gather their routing information from other routers, but they need a way to gather their information from other routers. So you need a standard, since there are a lot of different router vendors and you want every router to talk to every other router. But since the techniques evolve so do standards, so for routers to keep their information synchronized there are different standards, which use one of two different techniques: distance vector or link state.
protocol
Distance Vector
Link State
TCP IP
RIP, IGRIP
OSPF
OSI CLNP
IGRP
IS-IS
DECnet
DECnet phase IV
DECnet phase V
XNS IDP
RIP

AppleTalk DDP
RTMP

There are a lot more protocols but these are the most important ones. click on their names to learn more about them.
Distance Vector:

Distance Vector is also known as Bellman-Ford. The routers using this technique send their complete routing table to their neighbour routers. This can be an enormous amount network traffic.
Link State:

Also known as Shortes Path First sends only updates to the neighbour routers. This generates less traffic, but it uses more computation power on the router.