Paper Review:
Enabling Conferencing Applications on the Internet using an Overlay Multicast Architecture

Reviewer: Jie Zhou

Problem

IP Multicast has been studied for about 10 years. However, it is still not widely deployed due to the fundamental concerns related to scalability, and support for higher layer functionality like reliability and congestion control. Is IP the right layer to support multicast-routing related functionality?

Contribution

The authors of this paper have advocated an alternate multicast architecture, named End System Multicast. In this paper, via audio and video conferencing application, they explore the performance of End System Multicast in Internet environment which is dynamic, herogeneous and unpridictable.

Main Ideas

  • End System Multicast is a viable architecture for enabling performance demanding audio and video conferencing applications in Internet environment.
  • To achieve good performance, it is important to adapt to both latency and bandwidth while constructing overlays optimized for conferenceing applications.

    Critique

    I rate this paper at 4 (significant contribution), because implementing mulcast at application level is a fresh idea and this paper provides comprehensive analysis on the issues related with performance in real network, i.e Internet. The authors test different kinds of schemes for audio and video conferencing applications, which are intentionally running under dynamic and hetoenous environment. I feel their simulations and analysis are convincing.

    The authors interleave experiments with the various schemes, to make sure they are running at the same Internet condition. Since each experiment can only last for 20 minites every time, I doubt whether the incontinuity will affect the result of performance. Some schemes may converge to a good performance level only after enough long warm-up period.

    Lession

    Implementing multicast at application level has the two primary advantages: (1) it requires no network level support; (2) deployment is not an issue, as no change is required to network infrastructure.

    Open Questions

  • How to achieve shorter time scale adaptation targeted at extremely dynamic environments?
  • How to lower network costs for larger sized groups?