Data Networks, Local area networks

Topologies

The interconnections of computers in a local area network typically take one of the following forms :-

Fully interconnected - whilst giving high security, this is rarely used because :-
- expensive (of order n² cables and interfaces, where n is number of stations.
- difficult to maintain - addition of machine n+1 requires an extra n connections and interfaces.
Star network - all stations connect to a central hub which redistributes the data. Not seriosly affected by station failures or line breaks.
Ring network - stations each connected to two nearest neighbours to form a ring. Total failure occurs if a line break occurs. Station failure may be overcome.
Bus network - all stations connect to a single line (or bus). Not affected by station failure and forms disconnected sub-networks on line failure.

All topologies except fully connected share the communication medium using TDM so that all stations see all traffic on the network and cost increases as order n.

Physical vs logical topology

As we shall see, the logical view of the net often differs from the physical wiring layout. It is the logical topology which is prime importance.

Network interface hardware

As all stations in a LAN receive all traffic on the net, it is essential that processing of network traffic should not be done by the connected computers as :-

Processing all messages on all computers would severely restrict available cpu time for other work.
The network would run at the speed of the slowest machine. (networks often run faster than the computers to which they connect).

The work of processing network messages is handled by a separate attached processor on a network interface card.

Functions of network interface

The functions of the network interface card includes :-

Address recognition - identifying and accepting only messages addressed to the local computer.
Error checking - checking the frame CRC and rejecting frames containing errors.
Frame transmission - obtaining access to the network and transmitting frames sent by the local computer
Frame transfer - transferring frames to and from the local computers memory (using DMA) and interrupting the cpu to indicate completion.

As far as the attached computer is concerned, sending and receiving messages is similar to writing and reading data to a local file.

Signalling

Ethernet typically operates as a baseband network in which the data is transmitted as a digital wave form using Manachester encoding. The transmitting station sends an 8 octet preamble of alternating 1's and 0's (creating a square wave) which allows station receivers to synchronise their clocks.

As detection of voltage changes are easier than measuring actual voltage levels, Manchester encoding represent a 1 bit as an increasing voltage and a 0 as a falling voltage.

Physical constraints

Bus networks have physical limits which are affected by the physical characteristics of the medium and the logic of bus protocols :-

Signal attenuation - the bus length is limited by the attenuation of the signal.
Signal delay - the maximum bus length and minimum frame size are controlled by signal propogation delay.

In Ethernet, the maximum length of a bus segment is 500 meters and the minimum frame length including preamble is 72 octets.

Bus network characteristics

In bus networks, frames transmitted by stations propogate in both directions and are received by all other stations. The bus must be properly terminated or signals are reflected back along the bus.

Carrier-sense multiple access (CSMA)

Control of access to the bus is distributed between stations. A station wishing to transmit will listen to the medium before transmitting to make sure there is no traffic on the bus (listen before talk) and then transmits.

CSMA with collision detection (CSMA/CD)

If two stations start to transmit at approximately the same time, both signals will propogate and interfere with each other (a collision). To handle this, each station listens to the medium as it transmits to make sure the transmission is not corrupted (listen while talk). If a collision is detected, the transmitting party places a special jamming signal on the bus to make sure that the other party also knows about the problem and stops transmitting.

The transmitting station then waits for a short interval before attempting to send again. The length of the delay is based on a randomly sampled fraction of a fixed interval D. By selecting a random value, it is unlikely that the other party (which is doing the same) will choose the same interval.

Binary exponential backoff

If, after waiting a short time the station again suffers a collision, the processes is repeated but the randomly sampled interval D is doubled. This process is repeated until a succesful transmission occurs, doubling the sampled interval each time. After 16 attempts the station gives up and reports failure to the computer.

Frame format

Ethernet uses a frame format consisting of a fixed length header and a variable length payload. To ensure that collision detection works correctly, there is a minimum length for frames dictated by the technology used.

Ethernet (hardware) addresses

Ethernet addresses are 48 bits. Special addresses are selected for specific purposes:-

Broadcast address - an address of all 1 bits is reserved for broadcast messages which all stations accept.
Multicast addresses - starts with a 1 bit. Stations can be programmed to accept specific multicast addresses which allows subsets of stations to communicate.
Unicast addresses - all addresses starting with a 0 bit identify a unique station. Such addresses are allocated statically (i.e. fixed in the hardware by the NIC manufacturer).
Other ways in which LAN addresses can be allocated are :-
- Configurable - selectable by setting switches on the NIC or burning an EPROM.
- Dynamic - assigned by software when the station boots.
Both of these allow smaller addresses to be used, but in the case of dynamic assignment, stations have to attempt to send to an address to see if a response is made before chosing it as it's address.

Logical link control

Ethernet frame type field

During operation, a station may receive a variety of frames from different stations and for different purposes (as part of different protocols or conversations). For a station to be able to make sense of the payload, it needs to know what kind of information it contains. The type field serves this purpose.

A type field greater than 1500 (denary) indicates the payload type, smaller values indicate the frame length and indicate that the frame type is embedded at the start of the payload as LLC/SNAP.

LLC/SNAP

The logical link control, subnetwork attachment point header starts with AA AA 03 (hex) and is followed by the organization/frame type fields.

Thick Ethernet

The earlier forms of Ethernet used a thick coax cable which could not be bent round sharp curves and was expensive. Called 10base5, provided 10 Mbps using baseband with a 5 millimeter cable.

Physical arrangement

Because of the difficulty in bending the cable, transceivers which carried out the Manchester encoding and media-access control (CSMA/CD) are attached directly to the cable and the transceivers connect to the computer NIC's via an attachment unit interface (AUI) where address recognition, error detection and frame transfer is carried out.

To make computer attachment easier, connection multiplexors can be used to connect many computers to a single transceiver.

Thin Ethernet

A more popular form of Ethernet employs thinner coax cable. Called 10base2 it has a small diameter cable which can easily be strung between computers and is connected to stations via BNC connectors - making it susceptibel to disconnection. Removing a connector creates an unterminated end to the bus.

Twisted-pair Ethernet

Use of twisted-pair for Ethernet has become very popular due to its convenience and cost. Called 10baseT it uses CAT-5 twisted-pair wire and an RJ-45 connector (similar to a telephone connector).

It uses a hub instead of the normal bus to interconnect stations. Hubs typically have 8 ports and simply transmit all incoming data on one port to the other ports usually regenerating the signal in the process (i.e. they act like repeaters).

Other Ethernet Technologies

Ethernet technology has developed over the years and a number of different standards now exist, for example :-

10BaseFL - 2 strands of multimode optic fibre.
100BaseTX - 2 pair CAT-5 UTP.
1000BaseLX - Long-wavelength single-mode optic fibre.

Cable layouts

With these different technologies, buildings can be wired up in a variety of ways, e.g. :-

Token ring networks

There are numerous token ring technologies, but all rely on the passing of a special frame called a token to control access to the medium. Such networks do not suffer from collisions and perform better with high traffic loads.

IBM token ring

A station wishing to transmit :-

waits for a token to arrive and removes it from the ring
inserts a data frame onto the ring
receives the frame back - indicating succesful transmission or not
retransmits the token to the next station

A station not wishing to transmit simply passes the token and data frames on to the next station. The recipient of a frame copies the frame into a buffer and after checking the CRC transmits on to the next station, setting a bit to indicate successful receipt.

A station only transmits a single data frame before yielding the medium to ensure fair access to all parties. Only one frame (data or token) can exist on the ring at any given time.

Fibre Distributed Data Interconnect (FDDI)

An interesting variation of the IBM token ring involves a double fibre-optic ring in which data flows in opposite directions around each ring.

This provides greater reliability as a single station or segment failure results in automatic reconfiguration of the ring which maintains communication between all active stations.

Multiple failures result in a partitioning of the ring into sub-rings in a similar way to the partitioning experienced in bus networks.

Other ring & token-passing technologies

A number of other schemes exist of which two are notable :-

Cambridge slotted ring
This is a ring network which does not use a token but rather provides a number of frame slots (like a model train has a series of trucks) circulating the ring. A station wishing to transmit simply waits for an empty slot and inserts the data.
As a result there are a fixed number (more than 1) frames circulating at any given time.
Token bus
This is logically a token passing network as for the token ring, but the underlying network media is a bus and not a ring.

ATM switched network

A good example of a star topology is the ATM switch. This forms a central hub which connects to each station by a pair of optic fibre lines. As with all star networks, a station or link failure only affects the station concerned and the remainder of the network continues to execute. The switch however represents a single point of failure.

ATM switches operate at very high data rates and can be connected to each other. One switch forwarding frames on to another switch etc.

The Aloha net

Historically, this was the first wireless network developed (1960's) and was also the basis from which the Ethernet and other bus networks evolved.

Each station transmits its message to a central station which retransmits it to all stations on a different frequency. If the sending station hears the re-transmitted frame intact, it is assured of successful delivery.

If two stations transmissions overlap (by any amount), the central station will receive a collision and drops the frame.

The maximum theoretical traffic that can be sent is about 18% of the channel capacity (assuming equal-sized packets). This can be improved to about 30% by using slotted aloha which constrains stations to only start their transmissions at the beginning of slot times.

Aloha is used for communication between Earth stations and satellites. Either the satellite or an Earth station can act as the hub.

Modern wireless LAN's

The Aloha net uses centralised control. Modern wireless networks use ditributed media access control, typically with CSMA/CA - carrier-sense multiple access with collision avoidance.

Because of limitations in transmitter range, not all stations will hear every transmission. For instance, if computer 1 and computer 3, which are out of range of each other, attempt to communicate with computer 2 at the same time, computer 2 will receive a collision, but neither 1 nor 3 will detect it.

CSMA/CA

This works as follows :-

station 1 wishes to transmit to station 2 and sends a very short control message (RTS)
station 2 responds with a message saying go ahead (CTS) - received by all stations in range of station 2
station 1 transmits its frame - all other stations within range of station 2 keep quiet for a short period to prevent collisions at station 2.

Logical versus Physical topology

It should be clear now that the logical and physical topologies can differ and also differ from the cable layout, for instance :-

Token bus network running on 10BaseT
This is logically a token-passing ring network implemented on a bus network and using a wiring layout (stations connected to a hub) which forms a star.
Token ring using FDDI
A common way of implementing a ring network is to place the ring in a centrally located container and connect individual stations through wire pairs forming a star topology.
A particularly popular implementation of this uses an FDDI ring in the box so that it can automatically recover from individual station or link failures.

Network traffic analysis

Network interface cards can be put into so-called promiscuous mode by software instructions. In this mode they accept all frames regardless of the destination address (totally insecure).

Network analysers are dedicated computers which can be programmed to accept all traffic and analyse traffic in a variety of ways, e.g. :-

display or report on all frames from a specific station
generate statistics on all or specific types of frames
etc.