PACKET LOSS
Another Internet characteristic that is important to voice and video applications is packet loss. Two factors are involved: (a) how often packet loss occurs and (b) how many successive (contiguous) packets are affected. Packet loss is masked in a data application by using TCP to resend the lost TCP segments. Certainly, the loss of many segments and their retransmissions will affect the application's performance, but on the whole, the end-user data application is not concerned (or aware) of packet loss.
Packet loss is quite important in voice and video applications, since the loss may affect the outcome of the decoding process at the receiver, and may also be detected by the end-user's ears or eyes. Notwithstanding, today's voice coders can produce high-quality voice signals in the face of about 10 percent loss of the voice packets (G.723.1 as the example, see Chapter 4), if the packet losses are random and independent. G.723.1 compensates for this loss by using the previous packet to simulate the characteristics of the vocal signal that was in the lost packet.
Traffic loss in the Internet is bursty: large packet losses occur in a small number of bursts. This characteristic of internet behavior complicates the support of telephony, because packetized voice works best if the packet loss is random and independent.
The effect of packet loss can be alleviated somewhat by the use of forward error correction (FEC) schemes, and methods have been devised to compensate for loss bursts [BORE97].[1] These schemes add extra delay to the process, and may result in the loss of the packet because it is made available to the user in a time domain that is too late to be useful.
[1] [BORE97]. Borella, M. S. "Analysis of End-to-End Internet Packet Loss: Dependence and Asymmetry," IEEE Network, Reprint, 1997).
One FEC approach borrows from the tried-and-true mobile wireless technology: repeat the signal more than once. With mobile wireless systems, this operation interleaves successive copies of the coded voice image across multiple packets (slots in the mobile wireless terminology). With Internet telephony, experiments are underway to send copies of the packet 1 to n times. If one copy is lost, say copy n, it can be recovered from the other copies. However, this operation is only effective if a copy arrives safely, and therefore implies that one of the copies survives the burst error. If the copies are spaced-out too far in time to survive the error, they may arrive too late to be useful.
Order of Arrival of Packets
The subject of the order of the arrival of packets at the receiver is not of keen interest to the data application if it is supported by TCP, because TCP can re-order the TCP segments and present the traffic to the application in the correct order. TCP is not used for voice and video, so the order of packet arrival is an important subject to these applications. As of this writing, studies are underway to capture statistics and discover the incidences of misordered packet arrival. Several studies show that out-of-sequence arrival is not unusual.
Another factor is to note that Internet delays follow a diurnal cycle. During the hours of 8:00 a.m. to 6:00 p.m., delays are greater. For example, in the middle of the business day, delays are about 20 ms greater than in the evenings.
Hop Distance
Hop distance is a term used to describe the number of hops between a sender and a receiver. It is a critical aspect in internet telephony because more hops means more delay, and more variable delay. Hop distance must consider round-trip time (RTT), because of the interactive, real-time nature of telephone conversations. Figure 2-3 shows several aspects of hop distance and RTT, and its deleterious effect on the quality of voice applications.
The traffic is to be sent from a host on subnet 1 to a host on subnet 2. The IP datagrams must be processed by both hosts as well as all the routers on the path between the hosts. Let us assume the traffic traverses through the fewest number of hops (a common approach), which means the datagrams are processed by seven routers, numbered router 1 through router 7 in the figure. Thus, the datagrams are sent through nine hops. If the routers are not heavily loaded with traffic, then queuing delay will be short, and the delay at each router, while variable, will not create a major problem when the traffic arrives at the receiver. However, if traffic is heavy and/or if the routers are not performing their datagram forwarding operations efficiently, the accumulated and variable delay will result in the inability of the receiver to reconstitute the real-time voice signal into a high quality speech pattern.
We learned earlier that round-trip delay in the Internet rarely exceeds 200 ms. But we also know that the delay is highly variable. For example, a delay going through the same number of nodes might be 100 ms on one occasion; on another it may be 200 ms.
However, these studies focus only on the RTT in the network. Keep in mind that this RTT figure does not include analog-to-digital conversion, codec operations, or other factors that would increase the RTT.
Several studies also reveal that it is clear that geographical distance cannot be correlated to round-trip delay. Indeed in one study, a short distance of only 477 miles, but with a hop count of 21 resulted in a 500-ms round-trip delay. Therefore to emphasize, hop distance is a key factor in delay and geographical distance is less a factor.
Figure 2-3. Hop distance and round-trip delay (RTT)
The telephone network does not have these problems. First, the path between the talker and listener is fixed during the call set-up. Second, the circuit switches do not queue the traffic. Rather, the voice channels are time division multiplexed into DS0 channels (TDM slots) and sent directly from the input interface on the switch to a corresponding DS0 slot on a preconfigured output interface. The delay through the voice switch is miniscule (it is not even a factor), and fixed.
Thus, circuit switches provide fixed paths, as well as very low, fixed delay. In contrast, packet switches, such as routers, provide variable paths, and variable delay, which is sometimes a low delay and sometimes a lengthy delay.