Voice over IP (VoIP)

Quality of Service

Quality of service (QoS) is possibly the single most important feature to deploy to ensure a successful VoIP system. This section defines and describes why QoS is needed and explains how to configure and deploy a QoS solution using both the Modular QoS Command Line (MQC) and AutoQoS.


QoS Definition

QoS is defined as
The ability of the network to provide better or "special" service to a set of users and applications at the expense of other users and applications.

Voice and video traffic is very sensitive to delayed packets, lost packets, and variable delay (jitter). The effects of these problems manifest as choppy audio, missing sounds, echo, or unacceptably long pauses in the conversation that cause overlap, or one talker interrupting the other. QoS configurations provide bandwidth guar antees while minimizing delay and jitter for priority traffic like VoIP. They do so not by creating additional bandwidth, but by controlling how the available bandwidth is used by the different applications and protocols on the network. In effect, this often means that data applications and protocols are restricted from accessing bandwidth when VoIP traffic needs it. This does not have much of an impact on the data traffic, however, because it is generally not as delay or drop-sensitive as VoIP traffic.

The areas that QoS can address to improve and guarantee voice quality are the following:

• Bandwidth

• Delay (including delay variation or jitter)

• Packet loss

Bandwidth

A VoIP call follows a single path from end to end. That path may include a variety of LAN and WAN links. The slowest link represents the available bandwidth for the entire path—often referred to as a bottleneck because of the congestion the slow link can cause.

If congestion is occurring, there are several ways to fix the problem:

• Increase the bandwidth: If bandwidth is unlimited, congestion cannot occur. Realistically, however, increasing bandwidth is costly and is usually unnecessary if QoS is applied instead.
• Queuing: QoS employs advanced queuing strategies, which classify different traffic types and organize the classes into queues that are emptied in order of priority. The queuing strategies commonly used in Cisco Unified Communications include the following:

• Weighted fair queuing (WFQ): WFQ dynamically assigns bandwidth to traffic flows as they arrive at the router interface. No configuration is necessary; the strategy is enabled by default on router links of Tl speed and below. This strategy is not appropriate for VoIP because it does not provide a bandwidth guarantee for the voice traffic, but instead allocates bandwidth fairly based on flow sizes (hence the name). VoIP needs a Priority queue (PQ) that guarantees it the bandwidth it needs, at the expense of all other traffic.
  

• Class-based weighted fair queuing (CBWFQ): CBWFQ extends the WFQ algorithm to include user-defined classes for traffic. Instead of the router dynamically interpreting traffic flows and building queues for them, the admin classifies the traffic and assigns it to queues of configurable size and bandwidth allocation. There is still no priority queue, however, so CBWFQ is not appropriate for VoIP.

• Low-latency queuing (LLQ): LLQ extends the CBWFQ system with the addition of a PQ. The PQ is typically reserved for voice traffic, and if any packets show up in the PQ, all packets in the queue are immediately sent while packets of other traffic types are held in their respective queues. This is the preferred queuing method for VoIP networks.

Compression: Several strategies are available to make the data that needs to be sent smaller so that it consumes less bandwidth:

• Payload compression: By compacting the contents of a packet, the total size is somewhat reduced. This compression method does not affect the headers, which makes it appropriate for links that require the header to be readable to route the packet correctly (Frame Relay and ATM as examples).

• Link compression: On point-to-point links where the header information is not needed to route the packet, link compression may be used.

• Header compression: By specifying the use of compressed RTP (cRTP), the Layer 3 and 4 headers of a VoIP packet are dramatically reduced, from 40 to as little as 2 bytes. TCP header compression is also available for non-VoIP traffic using TCP transport.

Compression takes time and CPU resources, adding delay; this must be factored in to the decision of what strategies are appropriate for a given link.