LOCAL AREA NETWORKS
The subject of this book is premises data communications. I noted in Chapter 3 that there were three architectural alternatives to solving the connection problem posed in premises data communications. To this point our focus has been on only one of these approaches - the empirical techniques use to connect terminals to a 286/386 PC running UNIX/XENIX or some equivalent system. Another alternative is to use a Local Area Network (LAN). The employment of LANs will now be discussed in more detail.
8.1 WHY USE LANS?
Previously I made the following statement with regards to a LAN. "This alternative is the approach of choice when one has the freedom and the money to design their premises data communication system starting essentially with a `blank page’." This is a rather glib statement. The question arises as to why it is true?
Go back and take a look at the typical multi-user PC environment with its empirical connection solutions which was shown in Figure 2. Now take a look at the general pictorial of a LAN which is illustrated in Figure 25. The "cloud" represents the connecting media and all supporting protocols. There are many advantages which a LAN has over the data communications approach of Figure 2. I describe several below.
To begin with in Figure 2 one is only dealing with communications between the terminals and the multi-user computer. If there are printers or plotters or other equipment present they are essentially part of another data communications system with the multi-user PC at its center. In contrast as shown in Figure 25 a LAN allows full connectivity of all the data equipments connected to it. The terminals communicate with the computer through it. The computer can communicate with the printer or plotter through it. The terminals can even communicate with each other and with the printer or plotter through it. There is much more flexibility.
This advantage of a LAN in having full connectivity is often described in the context of it allowing peer-to-peer resource sharing.
The typical (UNIX-based) environment shown in Figure 2 does not have peer-to-peer capabilities. Rather, they designate the computers either as "clients" or "servers." A "client" is usually a computer at which a (human) end-user works. A "server" is a computer whose resources (e.g. printers, mass-storage devices, data files, program files, applications software) are available to all (human) end-users. Servers are often more powerful computers than the client computers. In the Figure 2 environment without peer-to-peer resource sharing the computer resources of a client are only available to the (human) end-user stationed at the client.
Figure 26: Data Equipments Connected By a LAN
However, a LAN has peer-to-peer resource sharing. All networked computers perform both client and server functions. This is the true meaning of full connectivity. It yields one of the major benefits which is file sharing among all the computers.
The multi-user computer system of Figure 2 will generally only make use of a data link level protocol to effect error control. LANs make use of packetization along with the higher levels of the OSI protocol stack. This leads to better error control as viewed from the user interfacing the application.
LANs are inherently faster. Data communications can be carried out at megabit per second (and above) speeds. They can support those applications which are becoming more and more common which need such speeds. Nowhere is this more evident than in the use if computer application packages using high resolution graphics. The multi-user computer system of Figure 2 with its 10’s of KBPS connections to terminals is a "time sharing system." Historically, it handled graphics by using "bit mapping." This was a relatively slow, cumbersome procedure for the UNIX/XENIX software and as a result graphics screens were painted "very slowly." Essentially, it was inappropriate for graphics. Recently, it has been improved by the development of the X-Terminals protocol. This replaces "bit mapping" with the transmission of "drawing descriptives." This makes the multi-user computer somewhat more appropriate for high resolution graphics. Nonetheless, the higher speed LANS have a significant advantage here.
8.2 LAN ARCHITECTURES
Work on the development of LANs began in the 1960’s. Since that time many different LAN architectures have been proposed, studied and implemented. However, at the present time only two architectures have survived to be of general interest, the ETHERNET BUS ARCHITECTURE and the TOKEN RING ARCHITECTURE. Neither of these has a clear advantage over the other. It is for that reason that both have been adopted as standards by the IEEE 802 committee. Other architectures may be of interest in special situations but these two are the only ones worth discussing of general interest. We do this below. Details are taken from [Ref.5] and [Ref. 6].
8.2.1 ETHERNET BUS ARCHITECTURE
This architecture had its origins in work done at XEROX’s Palo Alto Research Center (PARC) by Robert Metcalf in the early 1970’s. XEROX was later joined by DEC and INTEL in promoting ETHERNET as the coming LAN standard. Basically, Metcalf built upon work done by ARPA (now DARPA) and the University of Hawaii with a multiple access technique called ALOHA.
We can explain the operation of ETHERNET briefly with the aid of Figure 26. Here we see the data equipments which need to communicate all tapping onto transmission media (a cable) which we have labeled "Broadcast Channel-The ETHERNET BUS." The Bus Interface Units (BIUs) provide the essential interfacing between the data equipments and this channel-that is the transmit/receive capability and all needed intelligence. It is an essential feature of ETHERNET that by using the Broadcast Channel any data equipment can transmit to any other data equipment and any data equipment can listen to all transmissions on the channel whether intended for it or for some other data equipment user.
Now how does ETHERNET operate? It makes use of CSMA/CD - this is Carrier Sense Multiple Access/Collision Detection. The ETHERNET BUS - the connecting cable - is passive and can be used for broadcast type transmissions. Consider a specific data equipment unit, a terminal wanting to communicate with the computer. The terminal’s BIU before attempting to transmit a data packet onto the ETHERNET BUS first "listens" to determine if the BUS is idle-that is if there are no other packets from other data equipment units on the BUS. It senses the presence of a carrier on the BUS. An active BIU transmits its packet onto the BUS only if the BUS has been sensed idle. If the BUS is sensed busy then the BIU defers its transmission until the BUS becomes idle again. Due to propagation delays and carrier detection time, a collision may occur when a BIU senses an idle BUS and begins to transmit its packet while another data equipment’s BIU has already started transmitting a packet that has not yet propagated to this BIU. All BIU’s connected to the BUS have some means for "Collision Detection." When a collision occurs, all parties involved cease transmission and wait a random amount of time before initiating retransmission. If collision occurs again this random time wait is repeated but increased and increases at an exponential rate until the collision event disappears. This approach is called " exponential back off."
Figure 27: Ethernet Bus Architecture
8.2.2 TOKEN RING ARCHITECTURE
This architecture had its origins in work done in Great Britain. Its first known use in the United States was by Prime Computer in 1977. Since that time IBM has adapted this as its preferred LAN architecture.
The operation of the Token Ring Architecture can be explained with the help of Figure 27. In contrast to the ETHERNET BUS the Token Ring structure is a concatenation of point-to-point communications links arranged in a closed loop. Each link is terminated with an active repeater that detects a data packet on the in-bound link and re-transmits it on the outbound link. The detection, regeneration and all intelligence are carried out by the Ring Interface Unit (RIU) which interfaces the data equipment unit to the Ring. Basically, the ring transmission uses a token-passing access scheme. A special packet structure is called the "idle token." This circulates around the ring. When a connected data equipment user wants to transmit a data packet to some other user it may grab the "idle token," change it to a "busy state" and append its data packet to the busy token. At the end of the packet transmission, the data equipment unit issues another idle token. The Token Ring Architecture essentially behaves as a polling system. The most significant factor affecting the ring performance is the ring propagation time, the processing time for token recognition and regeneration at each RIU.
Figure 28: Token Ring Architecture
Ring structure is a concatenation of point-to-point communications links arranged in a closed loop. Each link is terminated with an active repeater that detects a data packet on the in-bound link and re-transmits it on the outbound link. The detection, regeneration and all intelligence are carried out by the Ring Interface Unit (RIU) which interfaces the data equipment unit to the Ring. Basically, the ring transmission uses a token-passing access scheme. A special packet structure is called the "idle token." This circulates around the ring. When a connected data equipment user wants to transmit a data packet to some other user it may grab the "idle token," change it to a "busy state" and append its data packet to the busy token. At the end of the packet transmission, the data equipment unit issues another idle token. The Token Ring Architecture essentially behaves as a polling system. The most significant factor affecting the ring performance is the ring propagation time, the processing time for token recognition and regeneration at each RIU.
8.2.3 COMPARING ETHERNET AND TOKEN RING
I have said that neither of these architectures has a clear advantage over the other. With ETHERNET the data equipment units do not have to be synchronized, they do not have to control their transmissions. This is a big advantage in implementation. It comes at the cost of collisions. But in lightly loaded networks collisions do not occur that often. Luckily, this is the situation most of the time. In heavily loaded situations, of course, long delays may occur and ETHERNET may degrade significantly. But these heavily loaded situations are almost never seen in the commercial environment.
With Token Ring transmissions are controlled. What is more, the maximum value of delay can be controlled by limiting the amount of time that a data equipment unit can take the idle token out of circulation. One never needs to suffer the infrequent long delays due to collisions that ETHERNET may experience.
Thus, in comparing the two architectures one needs to know where such infrequent long delays are tolerable and where they are not. They most likely are quite tolerable in the ordinary office environment where no catastrophe will result from the delay-just annoyance. Here, we would expect ETHERNET to be quite popular. On the
other hand if the LAN were being used in a factory environment to support process control or some type of automation an occasional long delay may well be catastrophic. Here, we would expect Token Ring to be the preferred architecture.
8.3 ETHERNET IMPLEMENTATION
I have noted that both ETHERNET and Token Ring are the LAN architectures which are mostly employed. Yet, to the present date ETHERNET appears to dominate as measured strictly by the amount of ETHERNET equipment in use. This dominance is expected to continue. It will be worth while to describe the evolution of ETHERNET implementations.
The first implementations of ETHERNET employed thick coaxial cable. Termed Thicknet it was defined by the 10 Base 5 standard. Figure 28 provides a pictorial illustration of this type of ETHERNET implementation. By comparing it to Figure 26 you can make the correspondence between the broadcast channel-the ETHERNET BUS and the two coaxial cable segments which are shown as vertical "pipes." Short segments of these pipes are joined by barrel connectors. Long jumps between these pipes required repeaters. The BIU function was carried out by a transceiver which was also referred to as the Media Access Unit (MAU). The Attachment Unit Interface (AUI) connected the data equipment to the MAU.
Thicknet delivered data across the network with a worst case bit error rate of 1 in a 100 million. A LAN implemented by Thicknet was also quite extensive. The maximum network cable length could exceed 8,000 feet. Implementation with Thicknet did require an external transceiver for signaling.
Figure 29: Thick Ethernet Network
Unfortunately, the thick coaxial cable was difficult to work with. As a result the second wave implementations of ETHERNET employed thin coaxial cable. Termed Thinnet it was defined by the 10 Base 2 standard. Figure 29 gives a pictorial illustration of this type of ETHERNET implementation. Thinnet had a bit error rate degraded somewhat relative to Thicknet. It also was not as extensive. The maximum network cable length was reduced to be of the order of 3,000 feet. Implementation with Thinnet did not require an external transceiver for signaling. Notice the absence of these in Figure 29.
The Thinnet realization of ETHERNET has recently given way to the use of Unshielded Twisted Pair cable (UTP) under the 10 Base T standard. The use of UTP has given ETHERNET deployments a new burst of growth. The cable used for the connecting medium is the same as that generally (although not always) used as the telephone cable in office buildings for the last 20 years. Thus, ETHERNET under 10 Base T can make use of an existing wiring plant and the task and cost of pulling new cable can be avoided.
ETHERNET under 10 Base T delivers data at the same worst case bit error rate of Thicknet, 1 in 100 million. However, a LAN implemented under 10 Base T has significant distance constraints.The UTP cable length must be of the order of 300 feet at its maximum. Implementation with 10 Base T does not require an external transceiver for signaling unless the adapter card is not 10 Base T compatible.
Figure 30: Thin Ethernet Network
The topology of an ETHERNET LAN under 10 Base T is also quite different from that of Thicknet and Thinnet. This topology is illustrated in Figure 30 taken from [Ref. 7]. The basic unit of a 10 Base T implementation is called the "Work Group." Data equipments are tied separately by UTP to a Multipoint Repeater or hub. This is a star topology, quite a bit different than the topologies for Thicknet or Thinnet. In fact, it is similar to the Multi-user PC Environment topology of Figure 2. Nonetheless, from a logical point of view this Work Group star topology has the required "Broadcast - ETHERNET BUS" property need for operation. Any data equipment unit can transmit to any other and also "listen" to all transmissions.
While ETHERNET under 10 Base T has the convenience of using UTP it also has the "annoyance" of having to employ 2 wire pairs to tie each data equipment unit to the hub. One pair is needed for transmission and another pair for reception.
The severe distance limitations on the Work Group distance also pose a problem. However, this can be solved by tying Work Groups together. This is indicated at the top of Figure 30. Depending upon the distance, Work Groups can be connected by either 10 Base T, 10 Base 2, 10 Base 5 or even a fiber optic cable.
Figure 31: Ethernet Operating As A 10 Base T Work Group
Figure 31 illustrates the components needed to connect to an ETHERNET LAN from the point of view of a data equipment unit "user." At the bottom of the figure is the applications software residing in the data equipment unit. This generates and is the ultimate recipient of the data to be communicated across the LAN. "On top of this" is the LAN operating system. Novell is probably the most popular example of this at present. The operating system carries out the packetization of data, complete with addressing. It establishes sessions between communicating data equipments. When collisions are sensed it handles the generation of exponential backoff procedures. "Above" the operating system is the adapter card. This carries out the CSMA/CD procedures. If it detects collisions it passes this information down to the operating system. With Thicknet only an external transceiver or MAU is also required as shown.
Figure 32: Connecting To An Ethernet LAN