Service Management

Beyond managing the CE router, you also need to consider managing additional services delivered as part of the VPN service. These additional services could include the following:

• Content distribution and web cache service

• Firewall

• Intrusion detection

• Voice services

• Denial of service (DoS) protection

• Internet access

When crafting SLAs, it is typical to create different metrics for different aspects of the VPN, including hardware, bandwidth, and any services provided. Clearly, the network connectivity has to be there for any service to run on top of it, so bandwidth availability has to be greater than or equal to service availability. Beyond simple availability, each service requires specific metrics or service characteristics relevant to the service operation. An example might be support required for Survivable Remote Site Telephony (SRST) to enhance voice service availability.

The negotiation of service management is typically specific to each enterprise's needs. It can be pivotal in deciding whether the service is continually managed by the enterprise or outsourced to the service provider.

Customer Reports and SLA Validation

Regular reports from the service provider form the basis of service validation and potentially rebates for nonperformance. Reports should provide measurement of metrics that are important to the enterprise, such as per-site performance on a per-class basis, per-site availability reports, and so on.

The reports available should be defined in the initial service agreement, with care taken to negotiate how availability, loss latency, and jitter are calculated.

The service provider's advertised performance for loss latency and jitter on its networks typically refers to PE-to-PE measurements that are averaged over groups of devices for a long period of time. As such, it is quite difficult to map those to what they mean for performance of a voice class at a particular location, for example. Care must be taken to ensure that the basis of SLA measurement and reporting of that measurement make sense from the enterprise perspective.

One of the issues discussed earlier in this section was that of the enterprise's remaining operational control over the CE device and the difficulty that poses in the provider offering CE-to-CE performance guarantees. One way to mitigate those difficulties is to request given loss latency and jitter characteristics, as long as the utilization on a per-class basis remains below a given levelsay, 70 or 80 percent. This protects the service provider, because if the enterprise puts too much traffic into the class, the provider is not responsible. However, if the class's performance is poor during times that the enterprise is not overloading the class, the provider is to blame, and penalties can be claimed.

How performance of the class is measured can be contentious. The recommended best practice is for active monitoring, such as the Cisco IP SLA probes, to be sent for each class. This lets per-class statistics for loss, latency, and jitter be collected. Typically an enterprise wants the provider to collect the results of these probes and format them into a regular report. However, it is common practice for enterprises to perform random verification of these reports with their own probe mechanisms.