| QoS is a nonnegotiable requirement for a successful VoIP deployment. With that in mind, this section discusses some recommendations for QoS and tuning techniques. Before you roll out QoS, assess your status. It is important to understand that QoS is not a replacement for VoIP-quality devices and adequate bandwidth. Only use QoS if, after adding all your anticipated VoIP traffic to your existing data network traffic, your network monitoring system is reporting that you are encountering congestion infrequently during the workday, yet your MOS declines unacceptably during those times. The recommended threshold for cumulative congestion per 8-hour day is less than 10 minutes. If you are seeing more congestion than that, go back and upgrade the network: Add more bandwidth, change your architecture, beef up your devices, and so on. Many different types of QoS and tuning techniques have been discussed in this chapter. The QoS techniques you use will depend a great deal on your network infrastructure and the other application traffic on the network already. Here are some guidelines and recommendations. Of the IP QoS techniques: -
RSVP is probably the wrong technique to use for VoIP traffic. However, consider applying it to long-running video streams that may be present in your network traffic. -
MPLS is best used by ISPs with large, complex networks. These networks may have dedicated paths through them, tuned for low delay, jitter, and loss—just what VoIP traffic needs. Here, the MPLS technique makes sense— it can give selected traffic dedicated paths through the backbone of a network. In fact, MPLS may be the preferred way for service providers to internally implement premium VoIP service. But you will still need a way to distinguish the VoIP traffic from the other application traffic at the edges of the network, so classifying based on port numbers, RTP headers, the DiffServ field, or packet sizes is still needed. -
Setting the DiffServ bits for VoIP traffic makes a lot of sense. Classify the traffic as close to its origin as possible. Other QoS techniques, such as CBWFQ, can use the DiffServ-marked packets to identify traffic classes. In addition, you need to configure the network devices, particularly the routers, to give traffic with different DiffServ markings different classes of handling. Making a cross-network configuration change for DiffServ requires some planning and good tools; approach it carefully. Of the link-layer QoS techniques: -
Use IEEE 802.1p/Q on Ethernet networks where the volume of VoIP traffic is likely to grow. Use 802.1p from softphones or IP phones to the switch. Plan to upgrade shared hubs, because they cannot provide the service required for VoIP traffic. Configure 802.1p/Q on the switches and create a separate VLAN for VoIP traffic. Use IP phones or softphone NICs that can set the 802.1p Priority field to a nonzero value. Configure the switches to provide priority to VoIP traffic that has the 802.1p Priority field set. -
Similarly, use ATM QoS on ATM links where the volume of VoIP traffic is likely to grow. Apply the CBR type of service to VoIP traffic where it is cost effective—because it is normally the most expensive service. Experiment to determine whether the less-expensive types of service are suitable for your deployment. -
On Frame Relay links, use admission control to limit the maximum number of concurrent VoIP calls allowed, then negotiate a CIR with your ISP to guarantee enough bandwidth for those calls. -
Use cRTP to double the VoIP bandwidth on relatively slow links. Use LFI to reduce delay and jitter on those same links. However, place high-horsepower devices at the ends of those links, because the extra processing for cRTP and LFI can consume a lot of CPU and add delay. In your routers: -
Use CBWFQ and LLQ to give VoIP traffic priority handling. Define the classifying technique, called the traffic match criteria, for VoIP using DiffServ bits. Use WRED to improve the handling of marked DiffServ traffic when congestion occurs. If you encounter congestion problems in the LLQ, you have to reserve more bandwidth for VoIP. -
If you have a network with a wide mix of traffic where VoIP is a small percentage, WFQ may help the VoIP traffic compete better with the other traffic. Be sure to test WFQ, though, as you deploy it, to make sure it really achieves the effect you want. Using traffic shapers: -
If there are clear concentration points in your network, it makes a lot of sense to use traffic shapers. You can have them look at many different attributes of the network traffic, and decide how much bandwidth to give to each different kind of flow. And traffic shaping is probably the easiest QoS method to set up. -
Calculate the maximum number of concurrent calls that can be supported with a good average MOS. Use CAC (gatekeepers) to limit the maximum number of concurrent VoIP calls, and then use a traffic shaper to guarantee enough bandwidth for those calls. In your IP phone configuration: -
Use the G.711 codec if you have sufficient bandwidth; use G.729 if bandwidth is limited but your overall MOS will still be high enough (that is, your delay, jitter, and packet loss levels are low). -
Don't use silence suppression unless bandwidth is really scarce, you have already chosen a low-speed codec (to conserve bandwidth), and silence suppression will give you enough breathing room to have a good MOS. -
Optimize your jitter buffer usage, depending on how much jitter you observe and how much fixed delay you can afford to add. |
