Document Links

Contents Introduction Key Planning Considerations Design Objectives Getting Started General Design Connectivity Issues for Ethernet Servers and Services Conclusion Appendix A: Standards Bodies Appendix B: Tools and Equipment Appendix C: Network Monitoring Tools References/Credits Illustrations Figure 1: Network Layers Figure 2: Network Topologies Figure 3: Hierarchical Star Network Tables Table 1: UTP Wires and Colors Printer Friendly Version Best Practices in Network Design (PDF file, 238 KB) Get Acrobat Reader

Best Practices in Network Design

General Design

Once you have the functional requirements and an idea of the network scale, you can begin the design. The network is comprised of layers, and decisions will need to be made on several things per layer.

Network Type: For a LAN, Ethernet at 10, 100 or 1,000 Mbps is the generally accepted standard. Some details involve switching vs. hubs and UTP vs. fiber. You will need to examine the bandwidth the network needs in order to support the applications. Multimedia (audio or video) demands more network resources and will consequently require a faster, higher-quality network, which usually costs more. For a CAN, the same network standards that are common in LANs are usually sufficient. For a MAN and WAN, however, networks based on Frame Relay or ATM are the most common choices.

Physical Network: The physical network covers everything from network cabling to faceplates and patch panels. The type of cabling you select depends on the type of network you need to support.

Network Communications Equipment: Hubs, switches ad routers connect the network parts together and control the flow of information.

Network Operating Systems (NOS): The operating system that supports the network. There are three predominant OSs: Microsoft Windows 2000, Novell Netware, and Unix. The OS runs on the server, and provides file and printer sharing and other network services such as DNS, DHCP, and web server.

Network Server Hardware: The physical machines that the NOS runs on. Generally, network servers are larger and more powerful than desktop machines; they also tend to have more RAM and hard disk space.

Backup Software and Hardware: Equipment and software to make backups of the servers and the user machines.

Client Machines: Consider the hardware (PC, Mac) and operating system (Mac OS, Windows 95/98/2000, Unix) that users will be running.

We will now take a more in-depth look at each of these areas.

Network Type

LAN

One of the first decisions network architects make in designing a network is the network type.

There are several options for a LAN, including Ethernet, Token Ring, ATM and FDDI. Only ATM, however, is a cell-based network; the rest are frame-based, with a variable length frame.

Token Ring networks are usually found in organizations with a large amount of IBM equipment. Token ring technology has been surpassed by Ethernet in both the number of installations and speed and is being replaced.

ATM follows a different approach than other network types and competes with Ethernet for backbone and high-performance connections. In an ATM network, data is broken into small, 53 byte cells which are switched over virtual circuits between devices. Most ATM networks operate at 155 Mbps, although different speeds, including 1.5 Mbps on a T-1, 25 Mbps on a UTP network, and 155 Mbps and 622 Mbps on optical fiber interfaces are available. ATM is considered both expensive and difficult to configure; its use is generally constrained to the backbone and to high-performance workstations, where the cost is more easily justified.

Ethernet is the current choice for most networks, and its wide use has prompted many vendors to build equipment to support Ethernet-based networks (which has driven the price down). Ethernet is based on CSMA/CD (Carrier Sense Multiple Access/Carrier Detect) as specified by IEEE 802.3. These rules describe how all devices on the network communicate. Each packet has an origin and destination address, and each computer opens only the packet destined for it. Mechanisms exist to sense if more than one device tries to use the network at the same time. If a collision occurs, the devices wait a random interval and re-transmit.

Ethernet operates at a variety of speeds: 10 Mbps, 100 Mbps, and 1,000 Mbps (also known as gigabit Ethernet or GigE). Adapter cards can generally only run at one speed. These cards must be matched with the hubs or switches whose ports run at the same speed. Some cards and adapter ports are auto sensing; that is, they can determine the capabilities of the other device and run at the best speed possible. Today's equipment senses between 10 and 100 Mbps. Very few 1,000 Mbps auto-sensing devices are currently available.

Ethernet Switches vs. Ethernet Hubs: A hub is a shared media device. All of the users connected to a hub share the bandwidth, and there is no isolation of broadcasts. Switches improve the overall performance, as each device is on its own network, which increases the bandwidth available to the end device, as well as minimizing the impact of errors or network failures.

Some switches allow for virtual LANs, or VLANs, to be built. This allows several networks (e.g., students, faculty, staff and lab networks) to be built sharing the same physical hardware. Without VLANs, each network would need its own switches/hubs to remain separate (for security and bandwidth control). With VLANs, each port on a switch can be assigned to a specific LAN, and then only traffic for that network is passed on that port.

FDDI: FDDI is a 100 Mbps fiber-based network that operates similarly to token ring, with the exception that FDDI is faster. FDDI (and CDDI, Copper Distributed Data Interface, a UTP-based version) is the same as token ring in terms of operation and functionality. FDDI equipment, like token ring, is not very common and tends to be expensive.

WAN

A WAN (or MAN) uses different networking systems than LANs. Although some providers can offer Ethernet over long distances, the predominant network types are based on ATM or Frame Relay.

ATM, discussed earlier, can be used to tie locations together. One of the key advantages to ATM is the ability to map different virtual circuits, called VCs, for different applications (i.e., one VC for data, one for video, one for voice, etc.) and control each VC independently (bandwidth, type). ATM is also used to terminate many small circuits (such as Frame Relay and DSL) onto a few high-speed circuits.

Frame Relay is similar to ATM in its use of virtual circuits. Frame Relay, however, is a frame-based service, where frames can be different lengths, which allows for efficient data transport. Frame Relay has difficulty with video and audio, as these are time-sensitive applications and do not do well competing for bandwidth with data frames. Frame services are widely available and are low in cost compared to most other network types. One issue to consider carefully is the actual network speed; you may want to order a circuit with more capacity than you need and then only order circuits set to smaller amounts (the CIR, or Committed Information Rate, specifies how much data is guaranteed to be transmitted; all other data is "best effort").

Other

When selected Ethernet devices, consider manageability. As your network may be scattered over large areas, the ability to remotely manage the devices and troubleshoot the network becomes more and more important. The Simple Network Management Protocol (SNMP) supports each device communicating to a central Network Management System (NMS) which keeps track of device status, configuration and current operating parameters.

Back to top