by Mark Sportack
Local area network topologies can be described using either a physical or a logical perspective. A physical topology describes the geometric arrangement of components that comprise the LAN. The topology is not a map of the network. It is a theoretical construct that graphically conveys the shape and structure of the LAN.
A logical topology describes the possible connections between pairs of networked endpoints that can communicate. This is useful in describing which endpoints can communicate with which other endpoints, and whether those pairs capable of communicating have a direct physical connection to each other. This chapter focuses only on physical topological descriptions.
There are three basic physical topologies: bus, ring, and star. Each basic topology is dictated by the LAN frame technology. For example, Ethernet networks, by definition, have historically used star topologies. The introduction of frame-level switching in LANs is changing this. Frame-switched LANs, regardless of frame type or access method, are topologically similar. Switched can now be added to the long-standing triad of basic LAN topologies as a distinct, fourth topology.
A bus topology, shown in Figure 5.1, features all networked nodes interconnected peer-to-peer using a single, open-ended cable. These ends must be terminated with a resistive load--that is, terminating resistors. This singe cable can support only a single channel. The cable is called the bus.
FIGURE 5.1. Typical bus topology.
The typical bus topology features a single cable, supported by no external electronics, that interconnects all networked nodes peer to peer. All connected devices listen to the bussed transmissions and accept those packets addressed to them. The lack of any external electronics, such as repeaters, makes bus LANs simple and inexpensive. The downside is that it also imposes severe limitations on distances, functionality, and scaleability.
This topology is impractical for all but the smallest of LANs. Consequently, today's commercially available LAN products that use a bus topology are inexpensive peer-to-peer networks that provide basic connectivity. These products are targeted at home and small office environments.
One exception to this was the IEEE's 802.4 Token Bus LAN specification. This technology was fairly robust and deterministic, and bore many similarities to a Token Ring LAN. The primary difference, obviously, was that Token Bus was implemented on a bus topology.
Token Bus found extremely limited market support. Its implementation tended to be limited to factory production lines. Technological refinement of other LAN technologies and topologies has made this sophisticated bus LAN obsolete.
The ring topology started out as a simple peer-to-peer LAN topology. Each networked workstation had two connections: one to each of its nearest neighbors (see Figure 5.2). The interconnection had to form a physical loop, or ring. Data was transmitted unidirectionally around the ring. Each workstation acted as a repeater, accepting and responding to packets addressed to it, and forwarding on the other packets to the next workstation "downstream."
FIGURE 5.2. Peer-to-peer ring topology.
The original LAN ring topology featured peer-to-peer connections between workstations. These connections had to be closed--that is, they had to form a ring. The benefit of such LANs was that response time was fairly predictable. The more devices there were in the ring, the longer the network delays. The drawback was that early ring networks could be completely disabled if one of the workstations failed.
These primitive rings were made obsolete by IBM's Token Ring, which was later standardized by the IEEE's 802.5 specification. Token Ring departed from the peer-to-peer interconnection in favor of a repeating hub. This eliminated the vulnerability of ring networks due to workstation failure by eliminating the peer-to-peer ring construction. Token Ring networks, despite their name, are implemented with a star topology and a circular access method, as shown in Figure 5.3.
FIGURE 5.3. Star-shaped ring topology.
LANs can be implemented in a star topology yet retain a circular access method. The Token Ring network illustrated in this figure demonstrates the virtual ring that is formed by the round-robin access method. The solid lines represent physical connections, and the dashed line represents the logical flow of regulated media access.
Functionally, the access token passes in a round-robin fashion among the networked endpoints. Thus, many people succumb to the temptation of describing Token Ring networks as having a logical ring topology, even though it is no longer a ring.
Star topology LANs have connections to networked devices that radiate out from a common point--that is, the hub, as shown in Figure 5.4. Unlike ring topologies, physical or virtual, each networked device in a star topology can access the media independently. These devices have to share the hub's available bandwidth. An example of a LAN with a star topology is Ethernet.
FIGURE 5.4. Star topology.
A small LAN with a star topology features connections that radiate out from a common point. Each connected device can initiate media access independent of the other connected devices.
Star topologies have become the dominant topology type in contemporary LANs. They are flexible, scaleable, and relatively inexpensive compared to more sophisticated LANs with strictly regulated access methods. Stars have all but made buses and rings obsolete in LAN topologies and have formed the basis for the final LAN topology: switched.
A switch is a multiport data link layer (OSI Reference Model Layer 2) device. A switch "learns" MAC addresses and stores them in an internal lookup table. Temporary, switched paths are created between the frame's originator and its intended recipient, and the frames are forwarded along that temporary path.
The typical LAN with a switched topology features multiple connections to a switching hub. (See Figure 5.5.) Each port, and the device it connects to, has its own dedicated bandwidth. Although originally switches forwarded frames, based upon the MAC address, technological advances are rapidly changing this. Switches are available today that can switch cells (a fixed-length, Layer 2 data-bearing structure). Switches can also be triggered by Layer 3 protocols, IP addresses, or even physical ports on the switching hub.
FIGURE 5.5. Switched topology.
Switches can improve the performance of a LAN in two important ways. First, they increase the aggregate bandwidth available throughout that network. For example, a switched Ethernet hub with 8 ports contains 8 separate collision domains of 10Mbps each, for an aggregate of 80Mbps of bandwidth.
The second way that switches improve LAN performance is by reducing the number of devices that are forced to share each segment of bandwidth. Each switch-delineated collision domain is inhabited by only two devices: the networked device and the port on the switching hub to which it connects. These are the only two devices that can compete for the 10Mbps of bandwidth on their segment. In networks that do not utilize a contention-based media access method, such as Token Ring or FDDI, the tokens circulate among a much smaller number of networked machines.
One area for concern with large switched implementations is that switches do not isolate broadcasts. They bolster performance solely by segmenting collision, not broadcast, domains. Excessive broadcast traffic can significantly and adversely impact LAN performance.
These four basic topologies are the building blocks of local area networking. They can be combined, extended, and implemented in a kaleidoscopic array of ways. The right topology for your LAN is the one that is best suited to your clients' particular network performance requirements. More likely than not, this ideal topology will be some combination of the basic topologies.
Complex topologies are extensions and/or combinations of basic physical topologies. Basic topologies, by themselves, are adequate for only very small LANs. The scaleability of the basic topologies is extremely limited. Complex topologies are formed from these building blocks to achieve a custom-fitted, scaleable topology.
The simplest of the complex topologies is developed by serially interconnecting all the hubs of a network, as shown in Figure 5.6. This is known as daisy chaining. This simple approach uses ports on existing hubs for interconnecting the hubs. Thus, no incremental cost is incurred during the development of such a backbone.
Small LANs can be scaled upward by daisy-chaining hubs together. Daisy chains are easily built and don't require any special administrator skills. Daisy chains were, historically, the interconnection method of choice for emerging, first-generation LANs.
The limits of daisy chaining can be discovered in a number of ways. LAN technology specifications, such as 802.3 Ethernet, dictate the maximum size of the LAN in terms of the number of hubs and/or repeaters that may be strung together in sequence. The distance limitations imposed by the physical layer, multiplied by the number of devices, dictate the maximum size of a LAN. This size is referred to as a maximum network diameter. Scaling beyond this diameter will adversely affect the normal functioning of that LAN. Maximum network diameters frequently limit the number of hubs that can be interconnected in this fashion. This is particularly true of contemporary high-performance LANs, such as Fast Ethernet, that place strict limitations on network diameter and the number of repeaters that can be strung together.
FIGURE 5.6. Daisy-chaining hubs.
Daisy-chaining networks that use a contention-based media access method can become problematic long before network diameter is compromised, however. Daisy chaining increases the number of connections, and therefore the number of devices, on a LAN. It does not increase aggregate bandwidth or segment collision domains. Daisy chaining simply increases the number of machines sharing the network's available bandwidth. Too many devices competing for the same amount of bandwidth can create collisions and quickly bring a LAN to its knees.
This topology is best left to LANs with less than a handful of hubs and little, if any, wide area networking.
Hierarchical topologies consist of more than one layer of hubs. Each layer serves a different network function. The bottom tier would be reserved for user station and server connectivity. Higher-level tiers provide aggregation of the user-level tier. A hierarchical arrangement is best suited for medium- to large-sized LANs that must be concerned with scaleability of the network and traffic aggregation.
Ring networks can be scaled up by interconnecting multiple rings in a hierarchical fashion, as shown in Figure 5.7. User station and server connectivity can be provided by as many limited size rings as is necessary to provide the required level of performance. A second-tier ring, either Token Ring or FDDI, can be used to interconnect all the user-level rings and to provide aggregated access to the WAN.
FIGURE 5.7. Hierarchical ring topology.
Small ring LANs can be scaled by interconnecting multiple rings hierarchically. In this figure, 16Mbps Token Ring (shown logically as a loop, rather than in its topologically correct star) is used to interconnect the user stations and FDDI loops are used for the servers and backbone tier.
Star topologies, too, can be implemented in hierarchical arrangements of multiple stars, as shown in Figure 5.8. Hierarchical stars can be implemented as a single collision domain or segmented into multiple collision domains using either switches or bridges.
FIGURE 5.8. Hierarchical star topology.
A hierarchical star topology uses one tier for user and server connectivity and the second tier as a backbone.
Overall network performance can be enhanced by not force-fitting all the functional requirements of the LAN into a single solution. Mixing multiple technologies is enabled by today's high-end switching hubs. New topologies can be introduced by inserting the appropriate circuit board into the high-bandwidth backplane. A hierarchical topology lends itself to such combination of topologies, as shown in Figure 5.9.
FIGURE 5.9. A hierarchical combination topology.
In this example of a hierarchical combination topology, an Asynchronous Transfer Mode (ATM) backbone is used to interconnect the user-level hubs. FDDI interconnects the "server farm" while Ethernet interconnects the user stations.
Topological variation can be an important way to optimize network performance for each of the various functional areas of a LAN. LANs contain four distinct functional areas: station connectivity, server connectivity, WAN connectivity, and backbone. Each may be best served by a different basic or complex topology.
The primary function of most LANs is station connectivity. Station connectivity tends to have the least stringent performance requirements of the LAN's functional areas. There are obvious exceptions to this, such as CAD/CAM workstations, desktop videoconferencing, and so on. In general, compromises in the cost and performance of this part of a LAN's technology and topology are less likely to adversely affect the network's performance.
Providing connectivity to machines that have divergent network performance requirements may require the use of multiple LAN technologies, as shown in Figure 5.10. Fortunately, many of today's hub manufacturers can support multiple technologies from the same hub chassis.
LANs provide basic connectivity to user stations and the peripherals that inhabit them. Differences in the network performance requirements of user station equipment can necessitate a mixed topology/technology solution.
FIGURE 5.10. Station connectivity LAN.
Servers tend to be much more robust than user workstations. Servers tend to be a point of traffic aggregation and must serve many clients and/or other servers. In the case of high-volume servers, this aggregation must be designed into a LAN's topology; otherwise, clients and servers suffer degraded network performance. Network connectivity to servers, typically, should also be more robust than station connectivity in terms of available bandwidth and robustness of access method.
LAN topologies can also be manipulated to accommodate the robust network performance requirements of servers and server clusters. In Figure 5.11, for example, a hierarchical combination topology is employed. The "server farm" is interconnected with a small FDDI loop, while the less-robust user stations are interconnected with Ethernet.
FIGURE 5.11. Server connectivity LAN.
A frequently overlooked aspect of a LAN's topology is its connection to the wide area network. In many cases, WAN connectivity is provided by a single connection from the backbone to a router, as shown in Figure 5.12.
FIGURE 5.12. WAN connectivity of the LAN.
The LAN's connection to the router that provides WAN connectivity is a crucial link in a building's overall LAN topology. Improper technology selection at this critical point can result in unacceptably deteriorated levels of performance for all traffic entering or exiting the building. LAN technologies that use a contention-based access method are highly inappropriate for this function.
Networks that support a high degree of WAN-to-LAN and LAN-to-WAN traffic benefit greatly from having the most robust connection possible in this aspect of their overall topology. The technology selected should be robust in terms of its nominal transmission rate and its access method. Contention-based technologies should be avoided at all costs. The use of a contention-based media, even on a dedicated switched port, may become problematic in high-usage networks. This is the bottleneck for all traffic coming into, and trying to get out of, the building.
A LAN's backbone is the portion of its facilities used to interconnect all the hubs. A backbone can be implemented in several topologies and with several different network components, as shown in Figure 5.13.
FIGURE 5.13. LAN backbone.
The LAN's backbone provides a critical function. It interconnects all the locally networked resources and, if applicable, the WAN. This logical depiction of a backbone can be implemented in a wide variety of ways.
Determining which backbone topology is correct for your LAN is not easy. Some options are easier to implement, very affordable, and easy to manage. Others can be more costly to acquire and operate. Another important difference lies in the scaleability of the various backbone topologies. Some are easy to scale, up to a point, and then require reinvestment to maintain acceptable levels of performance.
Each option must be examined individually, relative to your particular situation and requirements.
A serial backbone, shown in Figure 5.14, is nothing more than a series of hubs daisy-chained together. As described in the preceding section, this topology is inappropriate for all but the smallest of networks.
FIGURE 5.14. Serial backbone, also known as daisy chaining.
The hubs interconnecting users and servers may be serially connected to each other to form a primitive backbone. This, as previously mentioned, is what is known as daisy chaining.
A distributed backbone is a form of hierarchical topology that can be built by installing a backbone hub in a central location. A building's PBX room usually serves as the center of its wiring topology. Consequently, it is the ideal location for a distributed backbone hub. Connections from this hub are distributed to other hubs throughout the building, as shown in Figure 5.15.
FIGURE 5.15. A distributed backbone.
A distributed backbone can be developed by centrally locating the backbone hub. Connections are distributed from this hub to other hubs throughout the building. Unlike the serial backbone, this topology enables LANs to span large buildings without compromising maximum network diameters.
Distributing the backbone in this fashion requires an understanding of the building's wire topology and distance limitations of the various LAN media choices. In medium to large locations, the only viable option for implementing a distributed backbone will likely be fiber-optic cabling.
A collapsed backbone topology features a centralized router that interconnects all the LAN segments in a given building. The router effectively creates multiple collision and broadcast domains, thereby increasing the performance of each of the LAN segments.
Routers operate at Layer 3 of the OSI Reference Model. They are incapable of operating as quickly as hubs. Consequently, they can limit effective throughputs for any LAN traffic that originates on one LAN segment but terminates on another.
Collapsed backbones, like the one shown in Figure 5.16, also introduce a single point of failure in the LAN. This is not a fatal flaw. In fact, many of the other topologies also introduce a single point of failure into the LAN. Nevertheless, it must be considered when planning a network topology.
FIGURE 5.16. Collapsed backbone.
LAN segments can be interconnected by a router that functions as a collapsed backbone. This topology offers centralized control over the network, but introduces delays and a single point of failure.
An important consideration in collapsed backbone topologies is that user communities are seldom conveniently distributed throughout a building. It is more probable that multiple LAN segments will be needed for any given community. It is equally probable that multiple segments will exist in close proximity. Collapsed backbone topologies need to be carefully planned. Hastily or poorly constructed topologies will have adverse effects on network performance.
In some of the cases where collapsed backbones are untenable solutions, a modified version may prove ideal. This modification is known as the parallel backbone. The reasons for installing a parallel backbone are many. User communities may be widely dispersed throughout a building, some groups and/or applications may have stringent network security requirements, or high network availability may be required. Regardless of the reason, running parallel connections from a building's collapsed backbone router to the same telephone closet enables supporting multiple segments from each closet, as shown in Figure 5.17.
FIGURE 5.17. Parallel backbone topology.
The parallel backbone topology is a modification of the collapsed backbone. Multiple segments can be supported in the same telephone closet or equipment room. This marginally increases the cost of the network, but can increase the performance of each segment, and satisfy additional network criteria, like security.
Careful understanding of the performance requirements imposed by customers, stratified by LAN functional area, is the key to developing the ideal topology for any given set of user requirements. The potential combinations are limited only by one's imagination. Continued technological innovation will only increase the topological variety available to network designers.
Numerous other criteria, both technical and financial, are also factors in the selection of a LAN topology. The overall topology should be determined by customer performance requirements. These options should be used to refine and/or temper topological design decisions.
It doesn't take too much of an imagination to conjure up a network topology that can't be cost justified. Even large, well-funded network implementations have finite budgets. The implemented topology must balance cost against satisfaction of existing user requirements.
The ideal topology may prove impossible to implement for a number of reasons. The physical wire and distribution throughout a building may be inappropriate for the planned network. Rewiring might be cost prohibitive. Similarly, if your company has an extensive financial commitment to legacy technologies, it might not be feasible to implement an "ideal" network and topology. Lastly, the lack of adequate budgeting can quickly scale back network plans.
These are valid reasons for tempering an idealistic topology. Therefore, they should be examined and factored in before hardware is purchased.
It is foolish to design a network without first considering what is likely to occur in the foreseeable future. Innovations in network and computing technologies, changes in traffic volume and/or patterns, and a myriad of other factors could greatly alter the users' expectations for network performance in the future. The network and its topology must be flexible enough to accommodate expected future changes.
LAN topology is one of the most critical components of aggregate LAN performance. The four basic topologies--switched, star, ring, and bus--can be implemented in a dizzying array of variations and combinations. These combinations are not limited to just those that were presented in this chapter. Many of today's LAN technologies lend themselves quite readily to creative arrangement and combinations. It is important to understand the strengths and weaknesses of each topology relative to the LAN's desired performance and underlying technologies. These must be balanced against the realities of a building's physical layout, cable availability, cable paths, and even cable and wire types.
Ultimately, however, the successful topology is driven by the users' required performance levels and tempered with other considerations like cost, expected future growth, and technology limitations. The biggest challenge is translating the users' requirements into megabits per second (Mbps) and other network performance metrics.
© Copyright, Macmillan Computer Publishing. All rights reserved.
){kind=link}
){kind=link}
){kind=link}
){kind=link}
){kind=link}
){kind=link}
){kind=link}
){kind=link}
){kind=link}
){kind=link}
){kind=link}
){kind=link}
){kind=link}
){kind=link}
){kind=link}
){kind=link}
){kind=link}