Up-front rightsizing and capacity planning are critical aspects of VoIP, as many organizations have learned the hard way -- after VoIP deployment. "Rightsizing" refers to the provisioning of appropriately sized WAN links to support convergence of voice traffic onto the data network. In most cases, the current WAN link will need to be upgraded in order to support converged voice traffic. The question that must be answered is how much bandwidth is required for each WAN circuit, and this question cannot readily be answered without significant analysis of the current voice call patterns and volumes, as well as future call patterns and volumes using VoIP.
Bandwidth for voice: A deeper look
Traditional voice calls using Time Division Multiplexing (TDM) technology use 64 Kb per call. VoIP uses IP as the technology to enable transport of voice traffic over a data network. The voice data is encapsulated within an IP packet. The process of putting the voice data into the IP packet is called encoding. The method used to encode the traffic and the transport mechanism (ATM, Frame, Ethernet, etc.) can have a great impact on the size of the packet. Two common options for VoIP packet encapsulation are G.711 and G.729.
There are many encoding types, also called "codecs." You can see all of them in this codec summary chart from VoIPForo.com.
Both the codec and the transport the packet traverses must be considered when determining the amount of bandwidth utilized by one VoIP call. This is the key. You understand one call, and then you can perform an analysis of the call flows to understand how large the pipe should be. At this point, it is probably beneficial to become good friends with the PBX support resource.
For each of the sites that will utilize VoIP over the WAN, it is necessary to determine what is called the "busy hour call volume," which is the maximum number of concurrent calls at any one time for that site. The busy hour call volume will provide the maximum number of traditional TDM calls that the voice network can support. Once you know this number, rightsizing the WAN pipe for data becomes a mathematical exercise.
Let's use G.711 and G.729 encoding as examples.
In both G.711 and G.729, phones transmit 50 packets per second in each direction. The only difference between these two encoding schemes is the amount of voice data -- and thus the size of the packet.
The bandwidth used by G.711 can be calculated using the following formula:
Bandwidth = (50 packets/second x [packetsize] bytes/packet x 8 bits/byte)
In this case, the bandwidth would be full duplex. That is, this would be the bandwidth used in each direction. This is most realistic because in an IP telephony network, all links (links to phones, LAN uplinks, WAN circuits, etc.) are full duplex.
The following table displays the size of the packets used in various G.711 encoding scenarios:
| LAN | WAN | WAN with Header Compression |
| Layer 2 header | 18 bytes | 6 bytes | 6 bytes |
| IP/UDP/RTP header | 40 bytes | 40 bytes | 4 bytes |
| Payload (voice) | 160 bytes | 160 bytes | 160 bytes |
| Total size per packet | 218 bytes | 206 bytes | 170 bytes |
| Expected BW usage | 87.2 Kb/s | 82.4 Kb/s | 68 Kb/s |
The bandwidth used by G.729 can also be calculated using the formula:
Bandwidth = (50 packets/second x [packetsize] bytes/packet x 8 bits/byte)
Again, in this case, the bandwidth would be full duplex. The following table displays the size of the packets used in various G.729 encoding scenarios:
| LAN | WAN | WAN with Header Compression |
| Layer 2 header | 18 bytes | 6 bytes | 6 bytes |
| IP/UDP/RTP header | 40 bytes | 40 bytes | 4 bytes |
| Payload (voice) | 20 bytes | 20 bytes | 20 bytes |
| Total size per packet | 78 bytes | 66 bytes | 30 bytes |
| Expected BW usage | 31.2 Kb/s | 26.4 Kb/s | 12 Kb/s |
Let's assume that the busy hour call volume is 10 calls. So if you were to use G.711 encoding over a Layer 2 Frame Relay WAN network, the actual bandwidth required for one call would be 82.4 Kb/s and, for 10 calls, this would translate into 824 Kb/s of additional bandwidth needed just to put the voice calls onto the data network. As you can see, the encoding mechanism makes a huge difference. If you used G.729 encoding over the same WAN link, the bandwidth required would be 264 Kb/s (26.4 Kb/s x 10 calls).
This is the correct way to rightsize the WAN. It takes some time but ensures that you get it right the first time. You cannot afford to be oversubscribed on your WAN links when trying to push voice traffic. QoS can solve the problem of voice getting priority, but if the link is half the size it should be, something will get dropped.
About the author:
Robbie Harrell (CCIE#3873) is the National Practice Lead for Advanced Infrastructure Solutions for SBC Communications. He has more than 10 years of experience providing strategic, business and technical consulting services. Robbie lives in Atlanta and is a graduate of Clemson University. His background includes positions as a principal architect at International Network Services, Lucent, Frontway and Callisma.
