Summary

IP telephony is one of the three components of Cisco's AVVID framework. When implementing an IP telephony solution, you must consider network management, high availability, security, and QoS. QoS can guarantee necessary bandwidth and acceptable delay, jitter, and packet loss. For VoIP traffic, packet loss should be less than 1%, one-way delay less than 60ms per leg, and jitter less than 20ms. QoS does this through prioritization. Voice and video, transactional applications, and data transfers are typically prioritized as listed.

QoS includes classification, marking, forwarding, policing, queuing, scheduling, shaping, and dropping of traffic. FIFO lacks QoS. IntServ provides QoS on a connection-by-connection basis, whereas DiffServ provides QoS on a hop-by-hop basis. DiffServ uses CoS with IEEE 802.1Q/P frames and either DSCP or IP precedence in the IP TOS packet field. Classification is the process of grouping traffic into classes by using the class-map command. Traffic policies are defined within the policy-map command. You activate your policies with the service-policy command. Marking can be used so that other devices in the network know how to prioritize traffic. Queuing can then be used to implement your QoS policies.

WFQ is the default queuing method used on serial interfaces running at E1 speeds or less. It separates conversations into low and high priority based on Layer 3 and Layer 4 header information. PQ has four queues and the high queue always has precedence over the lower queues. Use the priority-list command to associate traffic with one of the four queues. CQ has 16 queues and processes them in a round-robin fashion. You can change the byte count for a queue to allow it to process more or less information when its turn arrives. Use the queue-list command to associate traffic with one of the queues. LLQ uses both PQ and CB-WFQ. WRRQ is the default queuing method on Cisco's Layer 3 switches. Use the priority command to specify an expedite queue. In WRRQ, there are four queues with weights assigned to them. The better the weight value, the more preference the queue is given.

If packets must be dropped, it is typically the tail end of the conversations that are dropped. WRED is used to drop packets before congestion becomes an issue, which is different from queuing. Conditioning of traffic shapes it to remove the "burstiness" from it, thereby reducing jitter problems.