Previous Table of Contents Next


The 10-Base-T specification permits 1024 nodes per collision domain and has a variety of rules, such as the 5-4-3-2-1 rule that governs node placement and installation. However, broadcast traffic and protocol selection greatly erode those guidelines. Table 2.2 notes the recommended maximum number of nodes per broadcast domain for the various common protocols on Ethernet technologies. Other physical media may not support the number of nodes reflected in the table.

TABLE 2.2 Recommended Maximum Number of Nodes per Broadcast Domain (Figures Based on Broadcast and Protocol)

Protocol Number of Nodes

AppleTalk 200 or less
NetBIOS 200 or less
IPX 500
IP (well designed) 1000


A number of companies have successfully designed networks well beyond these figures. These numbers are intended to provide a generic guideline that covers broadcasts and other limitations of the networking equipment.

Please note the “well-designed” IP guideline. This is consistent with a tuned non-broadcast-oriented installation. Windows (NetBIOS) installations typically show minor degradation at the 200-node level, although tuning will permit an increase in that number. Windows NT installations that utilize WINS as opposed to broadcast-based server discovery typically scale very well. When combining protocols, it is best to use the smaller number and include a factor for the added broadcasts and other traffic. For example, an installation with both Windows and Macintosh systems would best be kept to approximately 150 nodes. An installation with Novell and Unix might be capable of 400 nodes, although an analysis of RIP/SAP traffic and other criteria is likely warranted.


The 5-4-3-2-1 rule was used in the design of 10MB Ethernet networks with repeaters. It is not applicable with switches and faster Ethernet installations. The rule stated that Ethernet networks could have the following: five segments, four repeaters, three populated, two unpopulated, and one network. This rule was a guide to prevent collisions and contention problems that would pass through repeaters.

Trunking in Network Design

A powerful tool for the modern network designer is trunking technology, which combines multiple VLANs onto a single physical circuit. This design permits a single interface to support numerous networks—reducing costs and making more ports available for user connectivity. Trunks may be used between switches and routers, as shown in the following figures, or between switches. Switch-to-switch installations are more common, although this trend is changing. Designers should also note that trunking technology is available on network interface cards for server connections. This design may be used to provide a local presence from one server onto a number of subnets without using multiple NICs. Consider Figure 2.2, which illustrates a nontrunked VLAN installation.


FIGURE 2.2  Non-trunked VLAN installation

As the diagram shows, the designer must connect each VLAN to a separate router interface. Thus, for this five-VLAN model, the designer would need to purchase and connect five different links.

Figure 2.3 displays a trunked installation, which provides a single, 100MB Ethernet interface for all five VLANs. This design is commonly referred to as the “router on a stick” design. Were the non-trunked VLANs connected with 10MB interfaces, this design would clearly provide as much theoretical bandwidth.


FIGURE 2.3  Trunked VLAN installation

However, many administrators and designers would fret about taking five 100MB interfaces and reducing them to a single 100MB trunk. While their concern is clearly justified, each installation is different. Fortunately, there is a compromise solution that can provide ample bandwidth and retain some of the benefits found in trunking.

Cisco has introduced EtherChannel technologies into the switch and router platforms. This configuration disables the spanning tree and binds up to four links to provide four times the bandwidth to the trunk. This solution works well in practice for a number of reasons, including:

  It is rare for all VLANs to require bandwidth concurrently in production networks. This fact allows for substantial oversubscription of the trunk without providing underutilized bandwidth.
  EtherChannel links may continue to provide connectivity following a single link failure, which can be an additional benefit in fault-tolerant designs. Normally, this addresses potential port failures on the router.
  The creation of new VLANs frequently requires the designer to order hardware to support the VLAN. Extra hardware is not a factor when combining trunking with channeling.
  Newer network designs make use of multilayer switching—including Layer 3 path-selection switching. These technologies significantly reduce the number of packets requiring the router, as they are routed once and switched for subsequent packets.

EtherChannel technology is independent of trunking technology, and the two may be combined. The concept is that two or more channels may be used to provide additional bandwidth for a single VLAN or trunk—thus, the link between two switches could operate at up to 400Mbps full-duplex (bonding four 100Mbps full-duplex links). The following sections describe the various trunking protocols.

ISL

The Inter-Switch Link (ISL) protocol adds a 30-byte encapsulation header to each frame. This encapsulation tags the frame as belonging to a specific VLAN. ISL is proprietary to Cisco, and while other vendors (including Intel) have licensed the technology, it is slowly losing market share to the ratified IEEE 802.1q standard. ISL provides a great deal of information in its headers, including a second CRC in the encapsulation. ISL trunks can be deployed between routers and switches, switches and switches, and servers and switches or routers.


It is likely that Cisco will migrate away from the ISL protocol in favor of 802.1q. Designers should consider this factor when evaluating the protocol. Such a migration, should it occur, will likely take many years to come to fruition.


Previous Table of Contents Next