Chapter Summary

The chapter described the need for QoS mechanisms on a converged network (for example, a network that transports voice, video, and data). Cisco recommends that packets be classified and marked as close to the source as possible.

However, classification and marking used in isolation don't alter a packet's behavior. Other QoS mechanisms (for example, queuing, policing, and WRED) can examine a packet's marking and make forwarding or dropping decisions based on those markings.

Queuing mechanisms determine which packets stored in a queue get forwarded first and how much bandwidth those packets receive. Cisco routers typically take advantage of LLQ as their queuing mechanisms in a voice environment. Catalyst switches, however, often use WRR as their queuing mechanism of choice.

As a router's output queue begins to fill up, WRED can randomly discard lower-priority packets to prevent a queue from filling to capacity. While WRED can benefit many data applications, voice network designers typically don't use WRED for voice traffic classes because the packet drops caused by WRED's packet discards impact voice packets (which use UDP instead of TCP) differently than TCP-based data packets (which go into a TCP slow start).

Not all QoS mechanisms make a minimum amount of bandwidth available for traffic classes. Some QoS mechanisms, such as policing and shaping, limit the amount of bandwidth certain applications can use. Policing typically limits an application's bandwidth by dropping packets that exceed a specified bandwidth limit. Policing might alternately mark packets with a lower-priority marking, as opposed to dropping those packets. Shaping is a kinder and gentler approach, because shaping temporarily stores excess packets and attempts to transmit those packets at a later time, when sufficient bandwidth is available.

On lower-speed WAN links (that is, less than 768 kbps), Cisco recommends link efficiency mechanisms such as LFI and cRTP. LFI tools include MLP for Point to Point Protocol (PPP) links, FRF.12 for VoIPovFR links, and FRF.11 Annex C for VoFR links (that is, links that send voice traffic out a particular permanent virtual circuit instead of sending voice traffic to a remote IP address). Each of these LFI mechanisms fragment large data packets and interleave small voice packets in amongst the fragmented data packets. cRTP header compression takes a 40-byte voice header and logically compresses the header down to only 2 or 4 bytes in size, thus reducing the WAN bandwidth required to transmit a voice conversation.

Realize that these QoS mechanisms are not meant to be used in isolation. Rather, many QoS tools complement each other. For example, you might benefit from using LLQ, cRTP, and WRED, and shaping all on the same router interface.