Improving the performance of MANs and WANs is handled in the same way as improving LAN performance. You begin by checking the devices in the network, by upgrading the circuits between the computers, and by changing the demand placed on the network (Figure 9.17).
Improving Device Performance
In some cases, the key bottleneck in the network is not the circuits; it is the devices that provide access to the circuits (e.g., routers). One way to improve network performance is to upgrade the devices and computers that connect backbones to the WAN. Most devices are rated for their speed in converting input packets to output packets (called latency). Not all devices are created equal; some vendors produce devices with lower latencies than others.
Another strategy is examining the routing protocol, either static or dynamic. Dynamic routing will increase performance in networks that have many possible routes from one computer to another and in which message traffic is "bursty"—that is, in which traffic occurs in spurts, with many messages at one time, and few at others. But dynamic routing imposes an overhead cost by increasing network traffic. In some cases, the traffic and status information sent between computers accounts for more than 50 percent of all WAN message traffic. This is clearly a problem because it drastically reduces the amount of network capacity available for users’ messages. Dynamic routing should use no more than 10 to 20 percent of the network’s total capacity.
Improving Circuit Capacity
The first step is to analyze the message traffic in the network to find which circuits are approaching capacity. These circuits then can be upgraded to provide more capacity. Less-used circuits can be downgraded to save costs. A more sophisticated analysis involves examining why circuits are heavily used. For example, in Figure 9.3, the circuit from San Francisco to Vancouver may be heavily used, but much traffic on this circuit may not originate in San Francisco or be destined for Vancouver. It may, for example, be going from Los Angeles to Toronto, suggesting that adding a circuit here would improve performance to a greater extent than upgrading the San Francisco-to-Vancouver circuit.
Performance Checklist
Increase Computer and Device Performance
• Upgrade devices
• Change to a more appropriate routing protocol (either static or dynamic) Increase Circuit Capacity
• Analyze message traffic and upgrade to faster circuits where needed
• Check error rates
Reduce Network Demand
• Change user behavior
• Analyze network needs of all new systems
• Move data closer to users
Figure 9.17 Improving performance of metropolitan and local area networks
Gigabit Ethernet in the Netherlands
MANAGEMENT FOCUS
SURFnet is the national computer network for education and research in the Netherlands. Demand for network capacity had been rapidly growing as more and more students started using the Internet, so SURFnet began looking for a way to significantly upgrade its WAN that connects more than 50 universities, libraries, and research centers.
SURFnet considered implementing SONET or ATM OC-192, but felt that 10 Gbps Ethernet provided similardata rates, was more familiar to their customers, and was more scaleable. SURFnet has leased fiber from Amsterdam to major regional centers around the Netherlands (Figure 9.18). Each of these regional centers is a POP and in turn provides connections to other universities, libraries, and research centers in its region, often via a 1 Gbps or 100 Mbps Ethernet MAN or WAN. Sometimes SONET, ATM, or E-carrier services (the European equivalent to T carrier services) are used for the regional connections, depending upon the demand.
The capacity may be adequate for most traffic but not for meeting peak demand. One solution may be to add a circuit-switched or packet-switched service that is used only when demand exceeds circuit capacity. The use of a service as a backup for heavy traffic provides the best of both worlds. The lower-cost dedicated circuit is used constantly, and the backup service is used only when necessary to avoid poor response times.
Sometimes a shortage of capacity may be caused by a faulty circuit. As circuits deteriorate, the number of errors increases. As the error rate increases, throughput falls because more messages have to be retransmitted. Before installing new circuits, monitor the existing ones to ensure that they are operating properly or ask the common carrier to do it.
Reducing Network Demand
There are many ways to reduce network demand. One simple step is to require a network impact statement for all new application software developed or purchased by the organization. This focuses attention on the network impacts at an early stage in application development. Another simple approach is to use data compression techniques for all data in the network.
Another sometimes more difficult approach is to shift network usage from peak or high-cost times to lower-demand or lower-cost times. For example, the transmission of detailed sales and inventory reports from a retail store to headquarters could be done after the store closes.
Figure 9.18 The SURFnet gigabit Ethernet WAN
This takes advantage of off-peak rate charges and avoids interfering with transmissions requiring higher priority such as customer credit card authorizations.
The network can be redesigned to move data closer to the applications and people who use them. This also will reduce the amount of traffic in the network. Distributed database applications enable databases to be spread across several different computers. For example, instead of storing customer records in one central location, you could store them according to region.
