12-1 Quality of Service for Voice | Cisco Field Manual[c] Router Configuration

Many Quality of Service tools are available with the Cisco IOS software. Several of them are critical to the proper operation of a voice network. This section provides a brief overview of which QoS tools to apply at the appropriate locations in your network. For a complete description of the QoS tool configurations, refer to Chapter 10, "Quality of Service." In particular, Section 10-1 provides a comprehensive and modular approach toward using all the QoS tools available.

Voice traffic should be classified as close to the edge of the network as possible.
1. In a network with IP phones connected to Layer 2 switches, voice packets are automatically labeled with an IP Precedence value of 5 (critical) or an IP Differentiated Services Code Point (DSCP) value of EF. Because a phone is directly attached to a switch, the Precedence value can be trusted and allowed to pass into the network unchanged.
  
  NOTE
  
  For more information about IP Precedence, DSCP, Class of Service (CoS), and class-based packet marking with class maps, refer to Tables 10-1 and 10-2 and Section 10-1.
  
  When passing voice packets between the Layer 2 and Layer 3 domains, you should use a class-based packet marking to set the DSCP, IP Precedence, and Layer 2 CoS flags. VoIP RTP packets should have IP Precedence 5, DSCP EF, and CoS 5. VoIP control packets should have IP Precedence 3, DSCP AF31, and CoS 3. Here is an example of the commands to do this:
```
  class-map L3-L2-RTP   match ip dscp ef   match ip precedence 5   class-map L2-L3-RTP   match cos 5   class-map L3-L2-Control   match ip dscp af31   match ip precedence 3   class-map L2-L3-Control   match cos 3   policy-map output-L3-L2   class L3-L2-RTP   set cos 5   class L3-L2-Control   set cos 3   policy-map input-L2-L3   class L2-L3-RTP   set ip dscp ef   set ip precedence 5   class L2-L3-Control   set ip dscp af31   set ip precedence 3   interface FastEthernet 1/0   service-policy input input-L2-L3   service-policy output output-L3-L2  
```
2. For other voice IP packets entering a network through a router, the packets should be reclassified at the network edge. Any IP Precedence or DSCP values already set on incoming packets cannot be inherently trusted, because a user or application might try to set flags in all packets in order to get the best service.
  
  Use packet classification that looks for VoIP RTP, VoIP control, MGCP, and Cisco Skinny protocolsall used for voice traffic. Voice control traffic should have the IP Precedence reset to 3 or IP DSCP reset to 26. VoIP RTP traffic should be passed unchanged, although the IP Precedence should be 5 or the IP DSCP should be 46. All other traffic can have the IP Precedence or IP DSCP reset to 0 for best-effort transport. Here is an example of the classification commands to do this:
```
  ip access-list extended VoiceControl   remark Match Skinny protocol   permit tcp any any range 2000 2002   remark Match H.323 protocol   permit tcp any any eq 1720   permit tcp any any range 11000 11999   remark Match MGCP   permit udp any any eq 2427   class-map voice-control   match access-group VoiceControl   class-map rtp   match ip rtp 16384 16383   class-map others   match any   policy-map Classify-voice   class voice-control   set ip dscp 26   class rtp   set ip dscp 46   class others   set ip dscp 0   interface FastEthernet 1/0   service-policy input Classify-voice  
```
Use Low Latency Queuing (LLQ) with a strict priority queue for voice, where voice mixes with other traffic across a WAN link. Refer to the priority command described in Section 10-1.
Use Link Fragmentation and Interleaving (LFI) for efficient voice transmission over a WAN link with a bandwidth of less than 768 kbps. Voice packets must be sent out across a serial link at regular intervals, not more than every 10 milliseconds . LFI fragments large packets before transmission so that the serialization delay for each packet is minimized. Voice packets are then interleaved into the data stream. To choose the maximum fragment size for LFI, multiply the bandwidth of the WAN link (in kbps) by 10 milliseconds, and divide by 8 bits/byte. Table 12-1 lists the recommended LFI fragment sizes based on link bandwidth.

Table 12-1. Recommended LFI Fragment Sizes

Link Bandwidth	Recommended Fragment Size
56 kbps	70 bytes
64 kbps	80 bytes
128 kbps	160 bytes
256 kbps	320 bytes
512 kbps	640 bytes
768 kbps	960 bytes
Greater than 768 kbps	N/A

Provision the appropriate amount of bandwidth for voice calls. Each type of Codec uses a different sized packet, generated at a different rate. Table 12-2 provides typical bandwidth required per call. When provisioning the data circuit, use a slightly greater bandwidth per simultaneous call than the table shows, due to the size of the Layer 2 frame headers.

Table 12-2. Typical Bandwidth Needed for Voice Calls

Codec Type	Sampling Rate	Voice Payload (Bytes)	Packets Per Second	Bandwidth Per Call
G.711	20 ms	160	50	80 kbps
G.711	30 ms	240	33	53 kbps
G.729A	20 ms	20	50	24 kbps
G.729A	30 ms	30	33	16 kbps

Use Frame Relay Traffic Shaping (FRTS) for Frame Relay WAN links.
Use Call Admission Control to protect the available WAN link bandwidth. Admission Control can be performed by a Cisco CallManager or an H.323 gatekeeper.