Introduction This chapter introduces the four basic network topologies, or layouts: mesh, bus, star, and ring. Next we examine three methods for controlling access to the network media, or channel. We will look at polling for historical perspective, then move on to contention and token passing, and give some time to the Great Religious Debate about which is "better". The remainder of the chapter is devoted to one of the IEEE's networking standards: 802.2 Logical Linkg Control. Chapters Two and Three look at the methods, protocols, and technologies that inhabit the Physical and Data Link layers of the OSI Reference Model. Topologies Mesh Bus Ring Star Hybrid Channel Access Methods Polling Contention Token Passing Token Passing vs. Contention IEEE 802.2 (Logical Link Control) Conclusion Self Check
			[ I ] TOPOLOGIES The word topology means "a topographic study of a specific object, entity, place, etc." Essentially, it is a look at the lay of the land. In networks the term topology cen refer either to the physical layout of the network (how the cables are run from device to device) or the logical layout (how data flows from place to place). Frequently discussion of topologies is married to discussion of channel access methods (the rules for how data gets on the network). In this text I have separated topologies and channel access methods into two distinct topics. The simplest topology is a point-to-point connection: a single communications line running from one system to another. Point-to-point lines were used originally to connect one mainframe or minicomputer system to another. They are still used in the Wide Area Network (WAN) environment to connect networked locations together: one networked office building might be connected to another across town by a point-to-point line. Multipoint connections connect three or more nodes together on the same line. Large computer systems (mainframes and minis) use multipoint lines to connect a number of terminals on one line with a communications controller acting as the "master" on the line and all of the terminals operating as "slaves". In LAN systems virtually every network technology uses the multipoint concept -- multiple systems sharing one line -- to connect stations together. Mesh In a true mesh topology every node has a connection to every other node in the network. This confers some tremendous advantages. No node will ever encounter a busy channel when trying to communicate to another node. Multiple pathways exist from any node to any other node so that a failure of one or more links will not sever communication. The aggregate bandwidth of the network as a whole (the sum of the bandwidth of each individual line) is tremendous.
	One place where logical mesh networks are gaining popularity is in the arena of "commodity supercomuting" -- building networks of cheap PCs running in parallel to provid supercomputer performance at desktop PC prices. The Beowulf Project at NASA is one of the pioneers in this field.		Mesh networks are also tremendously wasteful and complex when they are larger than just a few nodes. As the number of nodes goes up, the number of connections increases almost exponentially. This translates into more and more cable and more and more network interfaces as each additional node is connected to the network. This mass of connections is difficult and expensive to maintain. Because of the nature of most network traffic, most of the links will remain idle most of the time, resulting in a huge waste of network bandwidth. Most large WANs and some MANs and campus LANs will implement some level of meshing to protect against loss of connectivity due to a cable or device failure. Mesh connections will be used to stitch the major points in the network together, will multipoint connections reaching out to the rest of the nodes of the network.
	The bus topology is used throughout computing. Expansion cards use the Peripheral Component Interconnect (PCI) bus to communicate with a PC's system board; many devices use the Small Computer System Interface (SCSI) bus to connect; Universal Serial Bus (USB) is being offered as the Next Big Thing to connect peripherals to PCs.		Bus The bus type network is one step further along the evolutionary path from a point to point connection. In a bus network all of the nodes are connected to a shared network media -- tradtionally coaxial cable. When a station transmits on the network, the signal propogates in both directions along the wire from the point of origin to the ends. Each station on the network can "see" the signal as it goes by and thus has the opportunity to receive the signal (if it is the intended recipient, based on the destination address in the header) by copying the signal into memory. At the ends of the bus terminators absorb the signal so that it is not reflected back into the network at the ends of the wire. (Termination problems are one of the most common types of failure for all bus-type topologies.) Bus networks use the least amount of cabling as the cable runs directly from one system to the next. Ethernet, ARCnet, and LocalTalk are network technologies that can be implemented with bus cabling. Ring In a ring network each station has two connections: one connecting the station to its "upstream" neighbor, the other connecting it to its "downstream" neighbor. Ring networks are loops made up of one-way point to point connections. Each station receives data from the station before it, and repeats that information to the station after it. Information flows around the ring from the originating station, through all of the other stations and back to the origin. Ring networks have the same efficiency of cabling as bus networks. IBM's Token-Ring (a.k.a. IEEE 802.5) and Fiber Distributed Data Interface (FDDI) are two ring-type networks. Both will be covered in Chapter Three. Star Most networks today are wired physically as a star. At the center is a hub with the other nodes of the network fanning out from the center. Hubs can be active, powered devices, or passive, unpowered devices. Of the commerically available networks only ARCnet and Token-Ring can operate with a passive hub; all others require an active hub.
	A typical hub installation in a wiring closet on a corporate network. Click to enlarge		Star networks use the most cable as each node has a "home run" back to the hub. However, this makes a star network much easier to troubleshoot than a bus or ring type network. If one station is not able to transmit or receive on the network, then the trouble lies somewhere on the path from that station to the hub. If all the stations cannot transmit or receive, the problem is probably with the hub itself. This is worthy of note: the hub in a star-wired network does represent a single point of failure. If the hub fails it will stop all traffic on the network. In contrast, with a bus or ring network a problem with a single node can bring down the entire network. Finding that one node as the cabling works its way from station to station can be a technician's nightmare. From a logical perspective, star networks pass data either as a bus or as a ring. This may seem to make no sense, but this is where the distinction between physical and logical topologies comes into play. With a bus network the physical and logical topologies are the same. With a star network, the physical topology is what we have described: network nodes each with a cable running back to a central point -- a hub. However, logically the mode of operation either may be like that of a bus or ring. When operating like a bus, a signal comes into the hub from a system and the hub sends the signal back out to all of the other workstations. The manner in which the signal propogates through the network is the same as that in a bus: from one point fanning out to many. Hybrid Hybrid topologies are just that -- mixtures of some or all of the common network topologies. A large corporation might have ring networks at one site, bus or star networks at another site, and use point-to-point lines to tie them both together. Most large networks have a mixture of technologies and topologies. Because so many large businesses operate as a collection of semi-independant business units, networking often starts in more than one unit, at different times with different vendors and different technologies, resulting in overall corporate networks with diverse topologies in use.
			[ II ] CHANNEL ACCESS METHODS Channel access methods are the rules of the road for the network as seen from the bottom two layers of the OSI Model. Other protocols and methods provide the rules for other, higher layers in the model. Channel Access Methods (CAMs) describe how networking systems put data on the network media, how low level errors are dealt with, and how the network polices itself. In this section I will detail three CAMs: polling, contention, and token passing. Polling Polling is not the simplest CAM, but it is the oldest. It was developed concurrently with the multipoint topology, at the time of the creation of the first large scale mainframe computing systems. It's understandable to think of a mainframe as a single monolithic computer, but in truth a mainframe is really a collection of computers performing different tasks in support of the whole. There is a CPU which organizes the entire entity and runs programs; there are controllers ("Storage Directors" in IBM parlance) that operate the disk drives ("DASD" or Direct Access Storage Device" in the IBM wold {as opposed to tape, which is a linear access sysem!}); and communications controllers ("FEPs" or Front End Processors) which handle the task of actually communicating with the terminals and printers. It is the FEP that I'll examine to explain polling. The Front End Processor (FEP) is the "boss" of all the terminals and printers in a mainframe system. All of these other devices function as "slaves" to this "master". In polling, the master queries each device in turn as to whether it has any data to transmit. If the answer is yes, the device is then permitted to transmit its data. If the answer is no, then the master moves on and asks, or polls the next slave device. The process is repeated continuously as long as the FEP is operating. Polling is not a particularly efficient CAM; a lot of time is wasted with the master asking each slave if it has anything to say. Polling is a method with a high degree of administrative overhead. However, polling is fairly economical to implement. Only the master device needs to have much of any intelligence. The other devices only need to have enough intelligence built in to be able to correctly handle being polled and to respond. This dichotomy of having an intelligent controller and relatively "stupid" devices is the origin of the phrase dumb terminal. Because of its high overhead, polling is not a practical method for running a network when efficiency and high throughput are desireable. Terminals and printers operate at data rates in the kilobit range where a few milliseconds lost to polling isn't all that much; modern data networks operate in the megabit and gigabit range where a millisecond is a lifetime. So, an efficient CAM is a must in high speed networks. Contention Polling represents the epitome of centralized control. No device can talk to any other device except the master, and then only when the master tells it that it may "speak". Contention-based CAMs, particularly the early implementations, are the diametric opposites of polling.
	ALOHA used radio -- an unbounded media. Because all of the transmitters used the same frequency the ALOHA net operated in the same fashion as a collection of devices sharing the same piece of bounded media (the same piece of cable). If each transmitter used a different freqency then each site would have to have a receiver for all of the other sites' transmitters, and you would have the same waste of hardware and bandwidth that you have with a true mesh network.		One of the early predecessors of Ethernet (a contention-based network technology) was called ALOHA Net. ALOHA was developed at (you guessed it) the University of Hawaii. It was designed for packet radio, and operated essentially as a free-for-all. If a station had something to transmit, it did. If it didn't get a response within a certain amount of time that station assumed that its transmission must have "collided" with another station's transmission, and it sent its broadcast again. There was no real control over who transmitted when; just a method for recovering from errors. Each station was in competition, or contention, with its peers for the chance to get a packet transmitted on a shared frequency. A pure contention network like ALOHA works well when there are only a few nodes in the network and traffic is very light. However, as soon as the number of stations or the volume of traffic increases above a bare minimum, the efficency of pure contention breaks down and the nodes spend more and more of their time retransmitting trying to get their data through. This prompted the development of more effective contention-based CAMs The next generation of contention-based CAM was called "Carrier Sense Multistation Access (CSMA). Under CSMA when a station wanted to transmit data on the network it first listened to the wire to see if any other transmission was going on (that's the Carrier Sense part of CSMA). If the wire was quiet, then the station went ahead with its transmission. If the wire was busy, the station waited and listened again. The listen and wait cycle would be repeated until the wire was quiet and the station could safely transmit. This method is similar to looking both ways before you cross the street. (Pure contention is akin to just plowing right into the road on the assumption that there is no oncoming traffic.) CSMA permitted many stations (the Multistation Access part) to use the same network wire without their transmissions constantly colliding with each other. However, CSMA had a shortcoming. It's possible for two stations to want to transmit at the same time, sense the wire at the same time, conclude that the network is idle, and simultaneously begin to send data onto the network. The result is a collision. CSMA lacked a mechanism for recognizing that such a collision had occurred and that a retransmission was necessary. As the number of systems and volume of data grows on a network CSMA loses efficency due to the increasing likelihood of collisions. CSMA was a great improvement over pure contention, but still not quite good enough. The third generation of contention-based CAMs brought forward CSMA/CD and CSMA/CA. CSMA/CD is Carrier Sense Multistation Access / Collision Detection, and CSMA/CA is Collision Avoidance. IEEE 802.3 (Ethernet) uses CSMA/CD, and Apple's LocalTalk uses CSMA/CA. CSMA/CD uses CSMA's method for sensing the wire before transmitting, but adds the capability for a station to recognize that a collision is occurring and to take corrective action. When a stations which is transmitting detects that its signal has collided with the signal from another station it immediately stops transmitting. It sends out a jamming signal for a set period of time to alert the other transmitting station that a collision has occurred. Both stations then "back off" for a set period of time plus a random element. After the back off period ends the stations will start the transmission process over again. The random element is inserted so that each station will wait in the back off state for a different length of time. This is a precaution against having two stations locked in a cycle of repetative collisions, which might come about if both waited for the exact same amount of time before retrying their broadcasts. Contention has a number of advantages. Contention-based protocols have very low administrative overhead. When data volumes are low to moderate contention is a highly efficient method for regulating traffic on the network. Contention is also relatively simple, and thus less expensive to implement. On the down side, contention-based protocols become less efficient as traffic levels become heavy. Also, the maximum time that a station might have to wait to gain access to the network wire is not guaranteed. We can only approximate the maximum wait time using the statistical discipline of probablility. Because of this, performance of contention-based network is said to be probabalistic. Probabalistic networks are not suited for some applications that demand consistant network performance and access times, such as certain industrial control applications and streaming high-quility video across the network.
			Token Passing If polling and contention define the extremes of total control and total anarchy, the token passing CAM exists somewhere in the middle of the spectrum somewhat on the side of control. Like contention all of the stations participate in the operation of the network. Yet, like polling there are strict rules governing which station can transmit and for how long. Token passing uses a "token" -- a small frame of data -- to grant a station permission to transmit. Whichever station has the token can put data out on the wire. When it is done transmitting it passes the token along to the next recipient. In the ring topology network shown below the token simply moves from upstream to downstream neighbor, looping continuously. IEEE 802.5 (Token-Ring) and FDDI use the token passing CAM with a logical ring topology. ARCnet uses a token passing CAM with a logical bus topology. When an ARCnet station is done transmitting it passes the token on to the station with the next highest station address. The station with the highest station address passes the token to the station with the lowest address so that the cycle will continue. System rules in the protocol specifications mandate how long a station may keep the token, how long it can transmit for, how to generate a new token if there isn't one circulating, etc. Token passing CAMs are said to be deterministic: unlike the functioning of contention-based CAMs which can only be described in terms of probabilities, you can (with a copy of the system rules, the number of nodes on the network, and the total length of all the cable in the network) determine for a certainty the maximum amount of time a station will have to wait between opportunities to transmit. This makes token passing a good choice for networks that carry time critical information, such as streaming video or machine control traffic.
			Token Passing vs. Contention One of the longest, most contentious debates among network professionals has been over which is superior: contention or token passing (specifically, Ethernet vs. Token-Ring). Zealots on both sides have spilled tremendous amounts of ink making their cases for what they see as the One True Network. Proponents of token passing argue that because there are no collisions with token-based CAMs that their networks have the best behavior under heavy load; that the determinism makes it easy to know what's going on. Advocates of contention point out that most networks operate under lighter load conditions where contention's low overhead translates into greater throughput and efficiency. They also point out the greater cost of token passing networks. Neither side is entirely right or wrong. Token passing LANs do provide greater throughput under heavy loads, and that throughput is sustained at a high level longer before falling off as load increases when compared to contention LANs. Yet contention-based LANs do have a distinct advantage when load is light: on a CSMA/CD network, a station is never stuck waiting for 100 other stations to decide that they don't have anything to transmit before it gets permission to go: if no one else is transmitting the station can start sending immediately, it doesn't have to wait for a token to come around.
	FDDI, the "other" token passing LAN, has not seen that much erosion in its market share due to the availability of Fast Ethernet. FDDI has always been a high cost, very high performance solution. Cheap Fast Ethernet can undercut the cost of FDDI but can't come close to matching its performance.		In practical terms, industry forces have decided which CAM has won. Ethernet, based on CSMA/CD, is the dominant network technology in the late 1990s. The 300 to 350% price differential between Ethernet and Token-Ring hardware, as well as the speed difference between Fast Ethernet (100Mbps) and 16Mbps Token-Ring have pretty much sealed Token-Ring's fate. The appearance of High Speed Token-Ring (HSTR) offering 100Mbps speeds in late 1997 was probably too little too late to revive corporate interest in Token-Ring. ARCnet was cheap, but at 2.5Mbps to was too slow to compete with Ethernet even before the development of Fast Ethernet. The ARCnet Trade Association's (ATA) release of 20Mbps ARCnet+ is a case of way too little, way too late. Offering a new 20Mbps network in a world with 100Mbps and Gigabit networking is somewhat akin to arriving at the scene of an accident with a BandAid just as one of the victims is dying from losing a leg.
			[ III ] IEEE 802.2 (LOGICAL LINK CONTROL) The purpose of this section is to give you an abridged description of the role of IEEE's 802.2 standard in network technology. There is a great deal of detail behind the functioning of 802.2 that I will omit in favor providing a clearer description of what 802.2 does relative to other standards. In its effort to make designing physical protocols easier, IEEE's 802 working committee decided to break the Data Link layer of the OSI Model into two sublayers, Logical Link Control (LLC) and Medium Access Control (MAC). The purpose of the LLC sublayer is to provide a consistant interface to the Network layer. The definition of the LLC sublayer is covered under the 802.2 standard. If you examine the chart above you will see that there are three different MAC sublayers shown, but only one LLC. All of the different MAC sublayers are designed to interface in the same way with the LLC sublayer. This means that a network designer only needs to write the code for the LLC once: it provides a consistent interface to the Network layer, and any number of MAC sublayers can "hook" into it and use its services (providing reliable transmission control for the Network layer). The three MAC sublayers shown are defined by the 802.3, 802.4, and 802.5 standards. They define a CSMA/CD network (Ethernet), a token passing bus network, and a token passing ring network (Token-Ring). Each of these MAC sublayers will be examined in the next chapter. [ IV ] CONCLUSION Chapter Two has presented three broad topic ares: network topologies, channel access methods, and one of the IEEE 802 standards -- 802.2. You should be able to diagram the four main topologies (bus, mesh, ring, and star), and describe how polling, contention, and token-passing operate. Use the Self Check section below to test your knowledge.
			[ V ] SELF CHECK Describe the weaknesses in a pure contention CAM, that drove the development of CSMA. What shortcoming in CSMA prompted the creation of CSMA/CD? How does a station get permission to transmit data in a token passing network? In a polling network? Under CSMA/CD? Which topology is most cost-effective in terms of the amount of cabling used? Which is least cost-effective in terms of the amount of network interfaces required? Star topologies are said to be easier to troubleshoot than physical bus or ring topologies. Why? You have a transmission problem on a bus network. You have checked all of the workstation connections and they all appear to be OK. Where would you look next? Why might Token-Ring be less efficient than Ethernet under light network loads? What layer of the OSI Model did the IEEE subdivide into LLC and MAC sublayers? ARCnet was never standardized under an IEEE 802 standard. Which 802 standard does ARCnet come closest to?
			[ CH 1 ] [ TOC ] [ CH 3 ] 1999, 2000 Shipman | Created 1-20-99 | Updated 2-4-00