Class of service

When you create a differentiated service policy or an inbound admission policy, you also create and use a class of service.

Differentiated service policies and Inbound admission policies use a class of service to group traffic into classes. Even though most of this happens through hardware, you control how you group traffic and what priority the traffic must receive.

As you carry out quality of service (QoS), you will first define policies. The policies determine the who, what, where, and when. Then you must assign a class of service to your policy. Classes of service are defined separately and might be reused by policies. When you define the class of service, you specify if it can be applied to outbound, inbound, or both policy types. If you select both (outbound and inbound), then a differentiated service policy and an inbound admission policy can use that class of service.

The settings within the class of service depend on whether the class of service is used for inbound, outbound, or both types of policies. When you create the class of service, you might encounter the following requirements:

Codepoint marking

QoS uses the recommended codepoints to assign per-hop behaviors to traffic. Routers and switches use these codepoints to give traffic priority levels. Your server cannot use these codepoints, because it does not act as a router. You must determine which codepoints to use, based on your individual network needs. Consider what applications are most important to you and what policies must be assigned higher priority. The most important thing is to be consistent with your markings, so that you get the results you expect. These codepoints will be a key part of differentiating different classes of traffic.

Traffic metering

QoS uses rate control limits to restrict traffic through your network. These limits are placed by setting the token bucket size, peak rate limit, and average rate limit. See Token bucket and bandwidth limits for more information about these specific values.

Out-of-profile traffic

The final portion of a class of service is out-of-profile handling. When you assign the rate control limits above, you set values to restrict traffic. When traffic exceeds these restrictions, the packets are considered out-of-profile. The information in a class of service tells the server whether to drop UDP traffic and reduce TCP congestion window, shape, or remark out-of-profile packets.

Drop UDP packets or reduce TCP congestion window: If you decide to drop and adjust out-of-profile packets, the UDP packets are dropped. However, the TCP congestion window is reduced so that data rate complies with the token bucket rate. The number of packets that can be sent into the network at any given time decreases, and as a result reduces congestion.

Delay (Shape): If you delay the out-of-profile packets, they are shaped to conform to your defined handling characteristics.

Re-mark with DiffServ codepoint: If you re-mark out-of-profile packets with a codepoint, they are reassigned a new codepoint. The packets are not throttled to meet your handling characteristics, just re-marked. When you assign these handling instructions in the wizard, click Help for more specific information.

Priority

You can prioritize the connections that are made to your server by different inbound admission control policies. This allows you to define the order in which completed connections are handled by your server. You can choose high, medium, low, or best effort.