Network Computingwww.networkcomputing.com

Other indicators such as congestion notifications (Forward Error Correction Notification [FECN] and Backward Error Correction Notification [BECN]) are not critical to the SLA. These may provide useful internal information to the carrier, or perhaps some indication of a brewing problem. However, they do not directly affect application performance, so you should not penalize your carrier for the presence of frames marked as such.

Use What the Carrier Can Give You Mo st carriers offer periodic usage reports, but they impart little knowledge about the network's operation and your use of it. The principal source of this data is the carrier switch and SNMP (RFC 1604) via data collected in one-minute, 15-minute or 24-hour intervals. While these may provide you with some idea of your busy hours and other utilization patterns, they don't provide sufficient detail to help you make hard decisions about your network topology and the size of your PVCs. Given these limitations, you shouldn't pay a premium for this data; you should negotiate for regular and ad hoc reports--both in hard copy and electronic versions--as a standard part of your frame relay service.

The typical carrier service report lists kilobytes and kiloframes sent and received, as well as those that are marked DE. Utilization reports list the highest 15-minute interval, and the pea k utilization percent and average utilization percent. They may also indicate the number of frames that are actually discarded. Disc ards fall into several types. Ingress discards happen at the input port to the carrier's switch, and indicate either switch overloading (the carrier's problem) or a burst above the Burst Excess level (you haven't contracted for enough CIR). The next type of discard may happen at the egress from the frame relay network, often because the port is oversubscribed (sum of PVCs is greater than port speed) and there is too much traffic to fit onto the access line.

Carriers such as LCI International, Intermedia Communications and WorldCom may provide an optional management and reporting station. However, these stations frequently cost several hundred dollars per month, and provide only switch-level statistics. Until carriers add value, such as PVC reconfiguration and trouble-ticketing (both now provided by LCI International), most organizations will find it difficult to justify the extra expense.

· Establish common performance standards, no matter how imperfect. SNMP is the easiest method of collecting p erformance data. SNMP excels at providing data at the device, port and interface levels. But the counters are too coarse for accurate CIR measurement, requiring frequent, network-intensive polls to come even close. Most devices don't have enough local storage of counter states to reduce the need for polling. This is where products that use RMON or proprietary protocols can excel. SNMP and standards-based RMON also fail to provide good application-level information, so you won't be able to tell much about the source of network utilization spikes.

Nonetheless, SNMP has its traditional place in fault and configuration management (with alarms, remote equipment installation and ongoing device tuning). You'll get information on physical line problems (RFC 1406 T1 MIB) and device configuration (DSU/CSU proprietary MIBs). Specific to frame relay, you'll see PVC status information, such as the number of frames sent and the number discarded or marked DE (see RFCs 1604 and 1315 for more detail).

· Think be yond performance. CIR, loss and latency are typical performance measures that should go into the Frame Relay SLA. Other metrics include network availability, errors and congestion, as well as service-oriented measurements such as mean time to repair and circuit provisioning lead time. The most important part of the SLA is establishing a set of metrics that both parties can agree on. These could include router-to-router pings, SNMP statistics from customer equipment or the carrier's switch, or from an application running on the customer network management station. What matters most is a common agreement on the measurement, from what points in the network it is taken, by whom and how often.

· If your applications are critical, invest in tools. Because SNMP can't measure your network's performance accurately enough, a new breed of frame relay network management tools has emerged. These systems place data-gathering probes directly on your network, either in the form of dedicated hardware (as with NetScou t Systems' NetScout Manager Plus) or by imbedding intelligence directly into CSU/DSUs (like Visual Networks' Visual UpTime). Others attempt to imbed smarter SNMP agents in the CSU/DSU (ADC Kentrox's DataSMART 680 Frame Monitoring DSUs), or extending traditional frame relay access device functionality (Paradyne Frame Relay Access Units).