Network Computingwww.networkcomputing.com

Tools and techniques exist to help you understand your WAN performance and ut ilization. Many don't require any investment on your part--except for the time that good negotiation and management require. We'll examine the strategies for getting the most out of your frame relay network, and the rules that should drive any service-level agreement (SLA) you sign with your carrier.

· Focus on the basics. Proper performance measurement doesn't start with sophisticated hardware and software, but rather with an examination of the network attributes that really matter to your application. For example, if your critical application is in image file transfer, you'll be most interested in monitoring the network's loss characteristics (such as dropped frames). If your application involves real-time audio traffic, then delay--and the consistency of that delay--is more important than loss.

Once you know what's truly important, you can adopt a strategy for capturin g network data. For example, loss (due to dropped and errored frames) can be accurately measured using the interface counters on your own equipment. Periodically poll these from your network management station, and use a standard reporting tool or spreadsheet to summarize performance.

Delay is a trickier matter, for a variety of reasons. Network delay can come from many sources; delay can vary from moment to moment; and, most tragically, carriers won't take complete ownership of the problem. Nearly all service provider guarantees are based on latency across the backbone, measured switch to switch across only the carrier's network. Thus, published guarantees serve principally to protect the carrier from assuming full responsibility end to end--across customer premise equipment, local loops and their own backbone as well.

· Establish link budgets for your applications. Suppose you establish a service objective of one-second response time to your users. After some analysis, you've determined that the processing at the ends of a transaction--workstation and server--can consume up to one-half of a second, and the delay arou nd your campus backbone can reach 200 milliseconds (ms). If remote network and router processing takes as much as 30 ms, that leaves less than 270 ms of round-trip delay for the WAN links.

· Know what you're measuring. A well-tuned frame relay network can be much more efficient and cost-effective than its leased-line equivalent, particularly as the number of nodes increases. A general rule: The burstier the traffic, the more difficult and costly tracking utilization becomes.

First, a quick review of key frame relay network parameters. Associated with each virtual circuit is the Committed Information Rate (CIR), the amount of data that the network will guarantee to deliver over a time interval. The CIR level is also the major determinant of the cost of the long-haul circuit, so it's the item that should be most accurately tracked. Unfortunately, accurate CIR measurement can be quite a trick.

CIR is calculated as the Committed Burst Size (Bc) that the network will deliver during a specific tim e interval (Tc). In other words, CIR=Bc/Tc. Collecting data as fast as Tc (often set to one second) is required for accurately verifying that you are really getting your CIR. The trouble is, most SNMP management platforms can't give you detail by the second, and typical carrier reports don't come even close.