Paper Review : A Reliable Multicast Framework for Light-Weight Sessions and Application Level Framing
Reviewer : Seh Leng Lim
This paper discusses extensions to Application Level Framing (ALF) and Light-Weight Sessions (LWS) protocol models to provide a framework for scalable reliable multicast (SRM) . The focus of the paper is on reliability rather than on congestion control.
The main contribution of the paper is its detailed description of the usage of SRM via the distributed white board ¡°wb¡± application and its performance simulations for its loss recovery algorithms. In accordance with the ALF principle that applications should have as much flexibility and functionality as possible, SRM is designed to meet only the minimal definition of reliable multicast by providing eventual delivery of all data to all the group members without enforcing any particular delivery order. SRM helps to enhance the multicast group concept by maximizing information and data sharing among all the members, and reduces overhead by making each member responsible for its own correct reception of all the data. SRM adheres to the design principle of TCP/IP by building reliability on an end-to-end basis and by dynamically adjusting their control parameters based on the observed performance within a session, therefore facilitating adaptation to a wide range of group sizes, topologies and link bandwidths while maintaining robust and high performance. SRM has the potential to support the proliferation of distributed Internet conferencing tools due to its lightweight overhead features.
The key main ideas expounded are :
(a) The idea of a minimal reliability definition of delivering all data to all group members, deferring more advanced functionality if needed to individual applications.
(b) The idea of using ¡°Naming in application data units (ADUs)¡± model for multicast, whereby all data has a unique, persistent name to support reliable multicast
(c) The idea of local recovery via sending repair request messages to a limited neighborhood scoped administratively, or by multicast group membership or by TTL distance
I think that the paper has a significant contribution (rating of 4) to the study of Internet multicast . The authors have described in interesting detail how a distributed whiteboard application called ¡°wb¡± is implemented on top of their SRM framework. Through their simulations, the authors have observed that the SRM loss recovery algorithms with fixed timer parameters give acceptable performance for sessions willing to tolerate a small number of duplicate requests and repairs and willing to accept a moderate request and repair delay. Thus, they also came up with an adaptive algorithm that adjusts the timer parameters as a function of both the delay and number of duplicate requests and repairs in recent loss recovery exchanges. However, their simulations involve studying the performance of their loss recovery algorithms for individual packet drops in random and bounded-degree trees, and does not involve realistic topologies or typical patterns of packet loss. As acknowledged by the authors, there is still work to be done in the area of scalable session messages, local recovery, and congestion control..
System and application builders may have a better appreciation from this paper of the difficulties involved in implementing multicast applications.