Performance Analysis of Voice over IP Networks

Enterprise Networks & Servers, May 2004 by Breit, Lou

While the idea of a single managed network consisting of data and voice traffic is appealing for many reasons, the need to ensure acceptable voice quality remains a primary concern.

The H.323 standard is designed to allow VoIP services to integrate with packet based networks. The components of H.323 are listed below.

* H.323 terminals or LAN endpoints.

* H.323 gateways which interface between the LAN and a circuit-switched network.

* A gatekeeper system that handles control functions.

* An MCU or Multipoint Control Unit that provides conferencing services between endpoints.

H.323 terminals implement voice transmission functions and include a CODEC (Compressor/Decompressor) that can send and receive packetized voice. A workstation running specialized software can function as an H.323 terminal. CODECS are available in various implementations, each of which carries a different processing overhead and voice quality. Voice channels occupy 64Kbps and use PCM or Pulse Code Modulation when carried over T1 links.

Overtime, advanced compression techniques were developed allowing a reduction in capacity usage but preserving voice quality. These compression techniques are incorporated within CODECS, and can be rated or compared by using four parameters.

Voice Data Compression: CODECS compress voice data down from 64Kbps to a specified bit rate. These rates can range from SKbps to as low as 5Kbps, though these numbers arc for audio streams only. Adding IP and Ethernet to the protocol mix will increase the data rate beyond the initial measurement.

Voice Quality: Some CODECS can compress voice without sacrificing quality, while others produce noticeable degradation.

Complexity: CPU resource requirements may increase if the CODEC design is complex.

Delay: The algorithms within each CODEC require a certain amount of speech be buffered before compression takes place. This delay can add to the total end-to-end delay within a conversation, and force users to engage in a half-duplex type of conversation.

Terminals also support a wide range of signaling functions used for call setup, tear down, and compression standards.

The H.323 gateway serves as the interface between the H.323 and non-H.323 or packet network. On one side, it maintains a connection to a traditional voice network, while the other side is configured to communicate to packet based components. Signalling messages are translated, and voice traffic is compressed and decompressed.

A gatekeeper system is not required as part of the H.323 network, but if installed it manages H.323 zones, a collection of logical devices in the same IP address subnet, and load balancing or redundant fail-over. Gatekeeper systems may also provide SNMP or management services, while limiting the number of signaled calls.

The MCU allows for conferencing between three or more terminals. The MC or Multipoint Controller handles the signaling necessary to set up and manage conferences, while the MP or Multipoint Processor forwards traffic to the appropriate endpoints.

While all of these components play an important role within the H.323 specification, none of them are categorized as distinct physical entities. Their functions may be combined into a single device, or deployed as individual hardware across the network.

Total quality of a voice conversation traveling over a packet network may be adversely affected by other variables beyond the choice of a CODEC. These include latency, jitter, and packet loss.

Latency: Unlike a file transfer or oneway music/audio transmission, a two-way telephone conversation is very sensitive to network delay or latency. Round trip delays exceeding 250ms are noticeable by most users, so a one-way delay of no more than 150ms is desirable. When that latency is exceeded, callers will typically begin to talk over each other and suffer through a poor quality conversation.

Network delays can be broken into two categories - fixed and variable. Fixed delays are constant and related to distance, while variable delays may result from changing network conditions. The end-toend delay between users is measured by traversing the entire network backbone and can be reduced be reducing the number of router hops between clients. Variable delays may depend on router utilization and circuit load, which can be reduced through proper management and analysis.

Another point of delay is the CODEC itself. The compression algorithms, while reducing bandwidth requirements, may add as much as 30 to 40ms of fixed delay.

Jitter: Jitter is calculated based on the difference in arrival time between successive packets. Two measurements are usually considered; the average inter-arrival time, and the standard deviation. On a well performing network, the average inter-arrival time will closely match the actual packet arrival times, and the standard deviation will be low reflecting a consistent arrival time. To compensate for changing network conditions, many voice gateways contain a jitter buffer. This buffer slows down the packet rate before compression/decompression, creating a smoothing effect for network packet flow and minimizing spikes and variations. However, the buffer can add significant delay but is configurable based on traffic conditions, and is sometimes dynamically adjusted by the controlling software.