Data networks are usually measured in terms of various metrics and in the traditional telephony setup, the voice quality is assigned a single number signifying its quality. VoIP is a combination of both. Mean Opinion Score (MOS) is a popular subjective measurement of voice quality, prescribed by the ITU. Various approaches have also been developed to measure call quality objectively. These include the Perceptual Speech Quality Measure (PSQM), Measuring Normalized Blocks (MNB), Perceptual Evaluation of Speech Quality (PESQ), Perceptual Analysis Measurement System (PAMS), etc.
PSQM and PAMS use digital signal processing algorithms that compare telephony signals with an analog reference signal. However, these techniques are not very well suited to test the quality of VoIP calls, they are unable to track and report on the issues of delay and jitter. An actual two-way conversation cannot be measured satisfactorily by these techniques and these are not scalable. Delay in conversation comprises the following components:
• Propagation delay: This delay is proportional to the speed of light and depends upon the physical distance between the two communicators.
• Transport delay: Transport delay occurs because of devices such as routers, firewalls, traffic shapers, etc. The delay can either be constant or vary with the traffic.
• Packetization delay: This is a function of the codec speeds. Low-speed codecs, such as the G.723, take around 67.5 milliseconds to convert analog signals into digital packets. The extra time is required because these packets have to compress the packets to reduce their size. High-speed codecs such as the G.711 can do the packetization in approximately one millisecond.
• Jitter buffer delay: A jitter buffer helps to minimize the variations in the arrival times of the voice datagrams. However, sometimes in the event of excessive delay, packets have to be discarded.
Audio signals degrade to a greater extent with low-speed codecs due to lossy compression. VoIP datagrams travel using RTP and are lost in the network either due to congestion or due to arriving late at the receiver’s buffer. Attempts at improving voice quality should consider factors such as the total one-way delay in both directions, delay variation, and packet loss in bursts.
VoIP quality is tested by conducting a VoIP Readiness Assessment, which has three stages. In the first stage, if the MOS value for a single VoIP communication is low, it implies that the data network needs to be upgraded. In the next stage, the MOS value is acquired for several call volumes at peak traffic. The third stage is to test the network performance by running concurrent calls with heavy background traffic. It is not often that router-based data networks deliver VoIP calls with toll-quality. Issues with jitter and packet loss have to be sorted out before a fair number of calls with good quality can be supported by the network.
The bottlenecks in a VoIP network can be removed in various ways including increasing bandwidth, replacing equipment, reconfiguring the network or changing its layout, etc.
Bandwidth consumption with VoIP calls depends upon the codecs used and is invariably greater than what is assumed. One can safely assume 160 kbps bandwidth consumption with a G.711 codec and around 50 kbps with the low-speed codecs. Bandwidth can be conserved by using the following techniques:
Compressed RTP headers: These help to conserve bandwidth by reducing the size of the RTP headers to around one-tenth of their original size. However, this technique increases latency and can add to the delay.
Silence Suppression: This helps to reduce the size of the payload by making use of the times in a conversation when both users are quiet.
RTP multiplexing: It conserves bandwidth by placing multiple packets of voice information in a single datagram. This reduces the number of IP/UDP headers required for each audio packet. However, this leads to delay as the data can be sent only when multiple packets have been generated. Also, even a single datagram lost implies the loss of several audio packets.
Call Admission: This helps you to manage the number of calls such that the bandwidth is not overloaded by having to support several concurrent conversations. This frees the bandwidth for other network activities.
Sometimes, it is possible to boost network performance by upgrading the existing infrastructure without having to invest in additional bandwidth. Upgrades can help to reduce latency and increase capacity as well. In a heavily trafficked LAN, switches are a better option to manage IP multicast traffic as compared to hubs. Increasing the RAM of a router can be an inexpensive upgrade. A hardware-based firewall has greater capacity than a software-based one and should be installed as it reduces transport delay.
Changing the network architecture is a major decision and should be considered under circumstances where a more direct route with fewer hops could be implemented thereby controlling transport losses and clustering of traffic patterns to explore the possibility of placing the servers closer to the clients, this will help in reducing the backbone traffic.
The QoS of network devices should be maintained at all times to ensure consistency of performance. The type of traffic dictates the nature of QoS technique to be applied. A network that does not have QoS standards treats all traffic with equal importance and is referred to as a “best effort” network. With a QoS standard in place, a network either reserves bandwidth for premium traffic or gives it priority in the event of congestion.
