Abstract

In recent years, the growth of the Internet has spawned a whole new generation of networked real–time multimedia applications, such as VoIP, video-conferencing, video on demand, music streaming, etc. which have very specific, and stringent, requirements in terms of network QoS. VoIP in particular has become very widespread lately, as attested by several commercial offerings, and software packages that implement it. It is well known that the Internet was not designed with real–time applications in mind, so the quality of these services tends to be very variable in a best–effort network context. Several techniques have been developed in order to improve the perceived quality. In this contribution, we extend our previous work and study the performance of one of those techniques, namely media-dependent Forward Error Correction. This error correction mechanism consists of piggybacking a compressed copy of the contents of packet $n$ in packet $n + i$ ($i$ being variable), so as to mitigate the effect of network losses on the quality of the conversation. To evaluate the impact of this technique on the perceived quality, we use a $./M/1/H$ 3–class queue to model the network, and study different scenarios to see how the increase in load produced by FEC affects the network state