Ethernet Versus Token Ring

For simple situations, the design of a network can be very simple; but selection of the wrong technology could lock you into a very expensive

disaster in the long term. So how should you design your network and select the technology to use? Certain things are fixed, such as the geography of your company. So the network has to fit the buildings. Usually you end up with a mix of technologies, usually defined as local area networks within the buildings and wide area networks between distant sites.

Network standards

If you read the computer press, you will see hundreds of branded network products. But essentially these fall into two camps, those that are designed to conform with the international standard for networking, the OSI Open Systems Interconnect model (OSI), and those that are pure proprietary networks. The latter encompass a range of low speed PC networks together with products for interconnecting peripherals such as terminals and printers. Because these are proprietary, by definition, they are heading rapidly into the obsolescence bin.

OSI specifies a number of interconnection technologies, i.e. the glue with which networks can be physically built.

X.25 is a wide area technology best known as packet switching. In the UK it is offered as a public data network service by British Telecom as PSS (Packet Switch Stream). Mercury Communications offers similar services.

IEEE 802.3 (or to give it its ISO title, IS 8802/3) is a standard derived from, and compatible with, the original Ethernet standard developed by DEC, Xerox and Intel, introduced in 1980. Ethernet technology dominates the local area network market place, with about 70standard used by a great number of computer system vendors, as well as the three developing companies.

IEEE 802.4 is a standard for a token-passing bus technology. The most well known implementation is the MAP (Manufacturing Automation Protocol) used in some factory environments.

IEEE 802.5 is the standard based on IBM's Token Ring technology, introduced in 1985.

Choosing the technology

The majority of networks are used to interconnect multiple sites, with many computer systems on each site. A single site can be regarded as a subset of this, although we would hope that all successful companies eventually grow into multiple sites! So the basic topology is one of local area networks interconnected with wide area links. Since the choice of wide area services is

defined by the public suppliers the major decision is which LAN technology to adopt.

IEEE 802.4 MAP style networks are really limited to specific factory applications, whereas the bulk of the market place is looking for networking for the commercial and office environment.

So the choice reduces to Ethernet or token ring. To compare these two technologies, we must look at specific areas, but first let us describe the basic technology.

Ethernet

Ethernet is a bus technology, with all devices attached to a single piece of wire. In its simplest form, this is an unbroken length of coaxial cable. The computers attach by clamping on to, but not breaking, this cable. This extremely simple design results in very high reliability, since the only common component is the cable.

Since all devices have equal access to the physical cable, there has to be an arbitration scheme. This scheme is called CSMA/CD (carrier sense multiple access with collision detection). All devices have equal access and wait for a clear period on the cable before transmitting. Where two or more stations start transmitting simultaneously, there will be a collision and both stations will wait a short random time before retransmitting.

Because devices all have access to the cable, only active devices actually participate in the network. Devices can be added or removed on a live network with no disruption, since the cable is not broken.

Token ring

IBM's Token Ring differs at a very basic level, in that the arbitration scheme used is that devices are not free to transmit when they wish: rather, they have to gain the right to transmit. This right is in the form of a token which is continually passed around the network, a continuous ring.

Each station on the network is an integral part of the cable, with all signals passing into and out of it. Bach station must take all incoming data and retransmit it to the next station. When the token is received by a station, it decides whether to use it or not.

To allow for errors and the reconfiguration of the ring, there must be a network master (or masters), which ensures that the token is kept circulating and restarts the network in the ease of failure or loss of the token. Token rings are extended by interconnection of separate rings, so as to limit the maximum number of stations, and hence the propagation delay, of the token.

LAN performance

This is the area which is promoted most basically, "My network is faster than yours" advertising. Like comparing computer systems, where all the argument revolves around MIPS and benchmarks, raw numbers are a poor indicator of system performance.

The basic numbers are that Ethernet propagates signals at 10Mbit/s and IBM Token Ring at 4Mbit/s, with 16Mbit/s available in the future. Some proprietary token passing systems clock signals at up to 10 Mbit/s. That is not the real story. Both networks support many attached systems, all of which may be trying to use the network for different things. So what performance can any one system get from a network and how many systems can simultaneously use the network?

The time taken for a data packet to reach its destination varies linearly with a number of factors, whilst the network is loaded below a certain threshold. Above that threshold network performance drops off dramatically, with maximum network throughput some way below the maximum network speed.

On an Ethernet, the propagation delay varies with the overall length of the network, but not with the number of attached nodes. Also, the network overhead varies with the size of packets, smaller packets carrying more overhead per byte of data. As the load rises, as expected, each station is more likely to have to wait, for the cable to be clear before transmitting; and there are more collisions, so the delay rises linearly with load. This continues until the knee is reached when multiple collisions are regularly occurring and main stations are waiting to transmit.

On a token ring, the propagation delay depends on the number of stations in the ring, since each station has to pass the token, whether or not it has any data to transmit. This leads to the situation where even if only two stations wish to be active on a ring, they have to suffer the propagation delay of token passing through all other stations. In a similar way to Ethernet, delay rises linearly with load until so many stations wish to be active that the token is taken at each station it reaches, and the time taken for the token to traverse the whole ring becomes large.

There is a solution. The solution for both Ethernet and token rings is to break the network and hence the work load into smaller sub-networks and to link the subnetworks with bridges. It is the performance of these bridges which will define the absolute performance of the network.

Bridges

Because the bridge learns the topology of the network, it is able to make decisions about network packets that arrive at it. Where the data destination is on the same side of the bridge as it arrives at, it can be discarded; hence the traffic is kept local to the subnetwork. This technique might be used to separate a file server and its client workstations from the rest of the network, thus keeping overall network load down. This technique works well because computer usage tends to follow human patterns, with people interchanging data and sharing resources much more frequently at, a departmental or work-group level than with the rest of the company.

Over time, as loading increases, bridges can be inserted at strategic points on the network to solve any load-related problems. It is obvious that the performance of the bridge is critical to the overall network performance, since the bridge must be able to process all the packets that arrive in real time and forward, with minimum delay, some subset of them.

Network growth

We have already considered the network's response to increasing load, hut what of its physical growth? What are the limits?

Л single Ethernet segment can support up to 1024 attached devices and this can he grown by_t extending the network with bridges to over 8000 Stations. There is no advised maximum attachment since adding stations physically does not affect network performance.

The station limit on token ring depends on the cable used. For the simplest twisted pair cable at 4Mbit/s the station limit is 72 stations, but configuration guidelines suggest about 30excessive delays in token passing. For higher quality cables the station limit is 260. Beyond these figures the ring must be grown by bridging to another, separate ring.

Ethernet's great advantage over token rings is that no station relies on any other. Stations can be added and taken away at will. In fact even new attachments can be made to the Ethernet cable, whilst the network is live, without affecting performance. On joining an Ethernet, a station may choose to broadcast a regular status message, from which other stations may build up a database of active stations, but there is no absolute need to do so to maintain network integrity.

On the other hand, on a token ring, the integrity of the ring must be maintained at all costs. Adding a station involves breaking the ring; and even if this is transient, the configuration databases must be updated before

the ring can continue processing. Similarly, if a station fails, the ring must bypass it physically to maintain the ring.

In summary, Ethernet has a don't care, open approach to reconfiguration and growth, whereas token ring has a closed and controlled method.

Notes:

local area network (Ian, LAN) локальная вычислительная сеть

(ЛВС)

wide area network глобальная вычислительная сеть

distant site отдаленный абонентский пункт

conform подчиняться

proprietary собственнический;

патентованное средство

encompass включать

obsolescence bin хлам

glue склеивать, соединять

derive from производить/выводить из

vendor торговец

token-passing bus technology шинная технология передачи

маркера

clamp смыкать

arbitration scheme схема управления доступом

к общей шине

CSMA/CD (carrier sense метод множественного доступа multiple access с контролем носителя

with collision detection) и обнаружением конфликтов

(МДКН/ОК)

participate take part in

live network действующая сеть

gain get

token маркер

propagation delay задержка на распространение

promote develop

benchmark оценка производительности

raw необработанный

clock синхронизировать

threshold порог

drop off уменьшаться

node узел

knee изгиб, колено

traverse пересекать

discard отбрасывать; отказываться от

twisted-pair cable кабель из витой пары проводов

avoid избегать

transient временный

bypass блокировать; шунтировать

3. Прочитайте текст. What new network topology have you learned about?