Review: A Reliable Multicast Framework for Light-Weight Sessions and Application Level Framing

The paper describes a reliable multicast framework for light-weight
sessions and application level framing, called SRM (Scalable Reliable
Multicast). SRM is designed to meet a minimal reliability definition of
eventual delivery of all the data to all the group members without
guaranteeing any order. It requires only the basic IP delivery model and
builds reliability on an end-to-end basis, without requiring any change
or special support from the underlying IP network. SRM uses
receiver-based reliability since single-point, sender-based control does
not adapt or scale well for mulicast delivery. Furthermore, SRM uses the
naming of application data units for the sender and receiver to describe
the progress of their communication. The authors list the assumptions
upon which SRM is based:
- IP mulicast datagram delivery is available.
- All data has unique, persistent names which always refer to the same
data. Also, the naming convention must be such that it allows a
hierarchy over the name space.
- Source ID's are persistent.
- All participants join the same mulitcast group; there is no distiction
between senders and receivers.
The authors then describe the SRM framework, including session messages
to report the sequence number state for active resources, a loss
recovery mechanism discussed in detail throughout the paper and short
sections on congestion control and network partitioning.
The rest of the paper discusses the resquest/repair algorithm used for
loss recovery and reliable data delivery, which depends on a set of
request and repair timer parameters. The authors describe the algorithm
and show simulations for 3 simple topologies (chain, star and
bounded-degree trees) showing that the performance of the algorthim
depends on the underlying network topologies. They also learn that the
paramaters that perform well in one case may not do so in another. As an
improvement, the authors describe a an adaptive algorithm that adjusts
the timer parameters in response to the past behaviour of the loss
recovery algortihm.
Finally, the paper discusses local loss recovery in order to reduce the
bandwidth costs of recovery in cases where the neighborhood affected by
the loss is small. The authors describe administrative scoping, separate
multicast groups and ttl-based scoping (for which they provide
simulations of two-step recovery).
The paper briefly discusses future work on scalable session messages,
local recovery, congestion control, and an SRK "toolkit".
The paper only discusses one aspect of SRM and leaves the rest as future
work. Although the request/repair is discussed in detail, I find their
simulations could have been more clear.
Received on Sun Oct 29 2006 - 00:14:18 EDT