9.8. Best Practices for Quality of Service When building or expanding a network for Voice over IP, there are many standards from which to choose, as you can see. Each has distinct characteristics that makes it good for certain situations. But, in the end, your success with QoS standards can be only as deep as your use of them is practical. Here's a list of best practices for ensuring Quality of Service on the network. -
Choose the right standards
-
-
Multiple QoS standards can complement one another, but don't overdo it. Using MPLS, RSVP, and DiffServ all on a 300-node network, while possible, wouldn't be a very practical idea -
Don't expect the QoS policies you establish to be supported by legacy routers and switches. -
If you are coordinating several network managers, departments, or divisions, be sure you have their buy-in and a mandate to enforce the QoS policy you establish. Easier said than done, especially in large organizations -
If your network traffic is more than 70% data to voice, use a packet prioritization technique like DiffServ or 802.1p -
If your network is large (hundreds of voice nodes) and relatively busy, use DiffServ -
If your network is especially congested or very large (thousands of voice nodes), use RSVP -
Use the standards correctly
-
-
In DiffServ setups, classify IP voice traffic (RTP streams) as EF (Expedited Forwarding) -
In 802.1p setups, classify IP voice traffic using IP precedence tag 5 -
Build the network to favor VoIP
-
-
Use 802.1q VLAN, and establish one or more separate VLANs for IP phones and/or VoIP servers -
To be on the safe side, assume every data link will need an additional 20 to 30 percent overhead bandwidth for call-signaling and -routing protocols (ARP, OSPF, etc.) aside from that normally used by the voice traffic itself -
If you are able to manage queuing on your IP routers, use low-latency queuing (LLQ) -
Use network links with faster speeds. Adding capacity is an acceptable, albeit inelegant and sometimes noneconomical, way of solving QoS issues -
Don't use slow (sub-128 kbps) links at all. -
Don't use wireless Ethernet for large workgroups of VoIP users -
Avoid running VoIP sessions through a VPN unless absolutely necessary, as VPNs have a lot of overhead -
Use digital PRI circuits or IP-based trunks rather than analog POTS or Centrex trunks to connect to the phone company. This can lessen the amount of digital/analog conversion and subsequent latency -
Use voice-coding techniques that enhance QoS
-
-
Use the G.711 codec as often as possible; it is the most resilient waveform codec with a negligible packaging delay -
Some codecs offer packet loss concealment (PLC). Use them to decrease perceived quality problems related to packet loss, but use them sparingly, as they can create latency -
To minimize latency, decrease the packet interval. Intervals can go as short as 10 ms, but 20 is typical. Decrease the use of jitter buffers by avoiding features like PLC, if possible -
Avoid transcoding if possible. If absolutely necessary, use IP endpoints and VoIP servers that support wireless codecs such as GSM -
If echo is a problem (especially across low-speed links), use codecs that support echo cancellation -
Maintain utilization
-
-
Keep total packet loss on each Ethernet segment below 1% -
Keep packet loss on T1s and other point-to-point connections at 0% -
Keep packet loss and jitter on frame-relay connections, VPNs, and other network clouds as low as possible. Negotiate an SLA with the service provider to enforce this -
Establish a service-level agreement
-
-
Before you begin building your VoIP system, if you have time, determine the MOS (mean opinion score) rating of all call paths on your current system. Use these scores as a minimum effort when selecting standards and equipment for your new VoIP system -
After the implementation, particularly in large corporate or carrier-class networks, establish an SLA between you and your users that provides an MOS expectation for every call path . Use this SLA and associated MOS scores as a metric to show your success |
