A Reliable Multicast Framework for Light-weight Sessions and Application Level Framing
Reviewer: Jie Zhou
Unlike unicast, multicast applications have widely different requirement for reliability. As a result, the "one size fits all" reliable multicast protocols incur huge overhead,
and cannot simultatiously meet the functionality, scalability and efficiency requirements.
This paper describes a scalable multicast framework, SRM, and investigates its performance dependence with underlying topology and operational environment.
To meet the different reliability requirements, multicast framework should include an application's semantics in the design of that application's protocol.
Receiver-based ADU model works better for multicast than sender-based sequence model, which is used in TCP.
An adaptive algorithm that uses the results of previous loss recovery events to adapt the control parameter can achieve shorter delay and fewer duplicate requests and repairs.
This paper gives detaied description of SRM, and provides comprehensive analysis of the underlying principles and its performance in different topologies. Besides substantial
simulation experiments, an implementation on wb application has been realized and runs well. So, I give it a rating of 4 (significant contribution).
Though the authors claim that the SRM framework is generally applicable to a wide variety of other applications, they have not given sufficient supporting arguments. As they admit,
their wb implementation is difficult to be extracted and re-used in other applications.
The best way to meet diverse application requirements is to leave as much functionality and flexibility as possible to the application.
How about using hierarchical approach for scalable session messages on larger groups?
Which mechanisms should be used to define local-recovery neighborhoods?
How should individual members determine whether to send requests with local or global scope?