Question

4. "... individual services often have conflicting QoS requirements with respect to bandwidth, delay, jitter and packet loss for the desired QoS." [Source: Nokia Siemens Networks Corporation, "QoS-enabled IF networks for assured multinlav QeE Using different applications as examples, discuss what this means

Answer 1

The Quality of Services or QoS aims to provide better and desired data traffic service by reducing the jitter, packet loss, and delay or latency. It is a technology that provides the desired level of bandwidth (data throughput capacity) in a traffic network. It can facilitate the selected network traffic by facilitating in terms of Frame Relay, Asynchronous Transfer Mode (ATM), Ethernet and 802.1 networks, SONET, IP-routed networks etc.

To achieve the desired level of performance the QoS uses many tools and techniques like the traffic shaping and jitter buffer which varies among the organization. For some institutions, the QoS has been set in SLA or the service-level agreement. In such a case, the network provider guarantees the desired level of satisfactory performance.

Due to congested network links, the switches and the routers start dropping packets giving rise to packet losses. If packet loss occurs in real-time audio or video calls, it results in jitters (like drift in time, congestion and route changes) in the communication session. As a consequence, the communication quality degrades. The QOS must reduce the time a packet takes from the sender to the destination or the latency. Ideally, it should be zero. Higher the latency larger the overlapped voice and echo in VOIP (Voice Over IP). An important QoS function is to maintain data throughput capacity or the transmission of a maximum amount of data within a specified time.

The packet loss can be masked using the Packet loss concealment or the PLC technique which is based on the type of codec. The G.711 is a simple waveform codec in which the PLC replays the last collected sample with an increase in attenuation at progressive repeats (Refer: G.711 Appendix I for PLC). Since the waveform is not likely to differ between adjacent samples, this method is able to conceal a packet loss to nearly 20 ms of samples. However, if a 30ms packetization interval is used in voice communication, the PLC may not be able to conceal the single packet loss effectively and may result in worse received voice quality with respect to a given packet loss probability. To tackle such issue, the PLCs use the sophisticated low bit-rate, frame-level codecs like G.723, and G.729. Such PLCs can conceal a packet loss to the extent of 30 to 40 ms with the desired receiver intelligibility. Nevertheless, due care is taken in designing the VOIP networks to maintain a 0% packet loss ideally, although the packet loss due to network failures or the L2bit errors can be tolerated.

The delay or latency is serious in case the degradation in voice quality is excessive. To tackle the latency issue in VOIP, the designer always rely on the ITU standard G.114 specified targets although, this standard is under revision currently. According to this standard, a one-way delay of 150ms from mouth to ear (end-to-end) is considered satisfactory for telephone services. However, the QOS designer must take due to concessions in latency budget to accommodate different network delay components such as scheduling delay due to congestion, access link serialization delay, the propagation delay through the backbone, service delay, etc. However, it is observed that a 200ms end-to-end delay can provide the desired Mean Opinion Score (MOS) in voice quality although the ideal 150ms end-to-end delay has been prescribed as a standard by ITU G.114. The delay or latency purely depends on the parties concerned and the MOS accepted by them. Certain parties or organization allows higher delay with a corresponding degradation in VOIP quality, they should consider these factors while taking the design decisions.

The jitter or play-out buffers tune and convert a network’s variable delays into fixed or constant delays at the receiver end. This way these buffers convert the arrival of the asynchronous packet into a synchronous stream. In the process, a trade-off has to be made between the probability of interrupted play-out due to the arrival of late packets and the delay. As such, the out-of-order or late packets are not considered. The choice of the jitter buffer should be made judiciously since either under or overflow jitter buffer results in degradation of voice quality. If it is set very large, it adds up the end-to-end delay resulting in scarcity of delay budget in the network. On the contrary, a very small jitter buffer may result in the underflow or overflow of the buffer. In an overflow case, it is essential for the codec to play out a sample due to the presence of an empty buffer. On the other hand, in an overflow state, the arrival of the subsequent packet may not be accommodated due to already full jitter buffer. Both jitter buffer underflows and overflows cause voice quality degradation. The issues of jitter can be solved using adaptive jitter buffers to tune it dynamically to a size of the acceptable value. It has been observed that a jitter more than 30ms can significantly degrade the voice quality; hence this value can be kept as a standard jitter target.

The Bandwidth or data throughput capacity the VoIP consumes in bits per second can be estimated by summing the payload of the VoIP samples in bytes to the 40-byte IP, RTP headers, and UDP, multiplying the result by eight (to get the results in bits), and then again multiplying the results by the packetization rate. The default packetization rate is taken as 50 packets per second (pps).

For example: In a G.711 VoIP codec, assume Packetization rate = 50 pps

Let the time taken to generate a new VoIP packet = 20ms

Payload of each packet of VoIP = 160 bytes

UDP =8bytes,

RTP headers= 12 bytes

So,       IP+UDP+RTP headers = 20 +8+ 12 bytes = 40 bytes

Length of the packet = 160 bytes+ 40bytes = 200 bytes

= 1600 bits [note: 200 bytes * 8= 1600 bits]

Bandwidth of VoIP = 1600 bits * 50 pps = 80kbps.

The voice bandwidth consumption for G.711 at 50pps that includes Layer 2 Overhead is 93 kbps for 802.1Q Ethernet, 84 kbps for PPP (Point-to-Point Protocol ), 86kbps for MLP (Multilink PPP), 84 kbps for Frame Relay with FRF.12 and 106 kbps for ATM. Similarly, it differs for G.711 at 33pps as well as for G.729A at 50pps and 33pps.

*Note the additional bytes added to the Layer 2 overhead are

• Considering the preambles, up to 32 bytes in 802.1Q Ethernet.
• Up to12 bytes due to PPP.
• Up to13 bytes in MLP.
• 4 bytes in Frame Relay.
• 8 bytes in Frame Relay with FRF.12
• Varying amounts of overhead in ATM based on the requirements of cell padding.