VoIP Performance
The motivation for running voice over IP networks is to eliminate the expense of maintaining separate voice and data networks. It sounds easy enough to run voice over IP network – just encapsulate digitized voice in IP packets and go. Digitizing and packetizing voice is fairly straightforward, but there’s one other key issue that is much tougher to deal with.
The key challenge in building converged networks is performance. Voice communications has much more stringent performance requirements than data communications. The best way to understand voice performance requirements is to analyze the traditional voice communications network – the public switched telephone network (PSTN). First, with the exception of most “last mile†copper loops, the PSTN is a digital network. Since human voice is analog, voice traffic must be digitized before it enters the network and then converted back to analog on the receiving end. Pairs of codecs (coder/decoders) at the endpoints perform the conversions between analog and digital signals. To provide high quality voice the codecs use a technique called Pulse Code Modulation (PCM) that samples analog voice every 125 microseconds (1/8,000 of a second) and digitally encodes each sample as an 8-bit code. Since 8,000 of these 8-bit samples must be transmitted every second, PCM requires 64 kbps of bandwidth for each call. To ensure the quality of each call, the PSTN uses multiplexing and circuit-switching technology to allocate a fixed 64 kbps channel for the duration of each call. Since the bandwidth required is always available, there is very little end-to-end latency, no jitter (variation in latency), and virtually no data loss.
Toll quality is the standard of comparison for VoIP because we all take the performance of the PSTN for granted and expect similar quality from any new technology that attempts to take its place.
Unlike the PSTN, IP networks use packet switching rather than circuit switching technology. Packet switching works well for data because it maximizes bandwidth utilization by allowing all users to dynamically share network bandwidth. The downside of IP’s dynamic resource sharing is that it provides only a best-effort delivery service which does not guarantee the performance levels of specific traffic flows such as voice conversations. To overcome these IP performance limitations enterprises are beginning to employ bandwidth management techniques such as prioritization to ensure that critical applications get the performance they need. But bandwidth management alone simply allocates bandwidth to critical applications at the expense of other applications, many of which are also important to the enterprise.
Similarly, just adding more bandwidth is usually ineffective because any additional bandwidth will be consumed by the most aggressive applications, not the most important ones. What is needed, particularly on WANs where bandwidth is scarce and expensive, is a combination of adequate bandwidth and the ability to manage that bandwidth. Since VoIP consumes predictable amounts of bandwidth for each call in progress one of the first tasks in planning for VoIP is to determine the amount of bandwidth needed for the number of active calls to be supported by the network.
The key challenge in building converged networks is performance. Voice communications has much more stringent performance requirements than data communications. The best way to understand voice performance requirements is to analyze the traditional voice communications network – the public switched telephone network (PSTN). First, with the exception of most “last mile†copper loops, the PSTN is a digital network. Since human voice is analog, voice traffic must be digitized before it enters the network and then converted back to analog on the receiving end. Pairs of codecs (coder/decoders) at the endpoints perform the conversions between analog and digital signals. To provide high quality voice the codecs use a technique called Pulse Code Modulation (PCM) that samples analog voice every 125 microseconds (1/8,000 of a second) and digitally encodes each sample as an 8-bit code. Since 8,000 of these 8-bit samples must be transmitted every second, PCM requires 64 kbps of bandwidth for each call. To ensure the quality of each call, the PSTN uses multiplexing and circuit-switching technology to allocate a fixed 64 kbps channel for the duration of each call. Since the bandwidth required is always available, there is very little end-to-end latency, no jitter (variation in latency), and virtually no data loss.
Toll quality is the standard of comparison for VoIP because we all take the performance of the PSTN for granted and expect similar quality from any new technology that attempts to take its place.
Voice over IP – Mixing Oil and Water?
To run voice traffic over IP networks it is first digitized and packetized. Digitized audio streams are transported between endpoints by the real-time protocol (RTP). RTP is a connection-oriented end-to-end protocol that is designed to transport delay-sensitive information. RTP identifies the encapsulated payload type and includes sequence numbers and time stamps that are use to synchronize real-time information flows. RTP uses the connectionless, unreliable user datagram protocol (UDP) transport protocols because retransmission of lost or corrupted data disrupts real time audio streams. Delivering high quality voice communications over IP networks is a challenge because these networks have none of the characteristics that enable the PSTN to provide toll quality voice service.Unlike the PSTN, IP networks use packet switching rather than circuit switching technology. Packet switching works well for data because it maximizes bandwidth utilization by allowing all users to dynamically share network bandwidth. The downside of IP’s dynamic resource sharing is that it provides only a best-effort delivery service which does not guarantee the performance levels of specific traffic flows such as voice conversations. To overcome these IP performance limitations enterprises are beginning to employ bandwidth management techniques such as prioritization to ensure that critical applications get the performance they need. But bandwidth management alone simply allocates bandwidth to critical applications at the expense of other applications, many of which are also important to the enterprise.
Similarly, just adding more bandwidth is usually ineffective because any additional bandwidth will be consumed by the most aggressive applications, not the most important ones. What is needed, particularly on WANs where bandwidth is scarce and expensive, is a combination of adequate bandwidth and the ability to manage that bandwidth. Since VoIP consumes predictable amounts of bandwidth for each call in progress one of the first tasks in planning for VoIP is to determine the amount of bandwidth needed for the number of active calls to be supported by the network.
