Paper Review: IP Multicast Channels: EXPRESS Support for Large-scale Single-source Applications

Reviewer: Mark Meras (mm446)

Main Contribution

This paper by Holbrook and Cheriton proposes a new extension to the IP multicast model called EXPRESS. EXPRESS creates a channel model for delivering large amounts of data from one source to many subscribers.

Key Ideas

IP multicast is currently based on a group model of communication in which a set of hosts are grouped together into a single address. This approach has several problems:
- It violates the current ISP billing models
- It provides no indication of group size
- It does not provide a mechanism to restrict membership
- The group model allocates world-wide unique addresses, of which there are only 256 million
- It does not scale well
A better approach can be used, the so-called EXPRESS (EXPlicity REquested Single-Source) model. The EXPRESS model provides for channels that allow a pair (S,E) to be subscribed to by many subscribers, where S is the source (host) of the channel. Thus, the EXPRESS model capitalizes on the one-to-many nature of most IP multicast applications.
A new protocol, ECMP, is proposed to allow for subscriptions and counting. Reverse-path forwarding is used to keep the dynamic tree for IP multicast. ECMP sets up the Forwarding Information Base (FIB) entries at the routers that are used to look up the (S,E) pairs for multicast distribution.
A session relay approach can be used for almost single source multicasting, such as distance learning. In this application, one host is the relay to which everyone is subscribed. To send a message to all subscribers, a host simply sends it to the relay to be forwarded. Session relay is very easy to implement using the EXPRESS model.

Critique of Contribution

The new EXPRESS model seems to be a very important contribution to the field of IP multicasting. It uses an existing infrastructure (IP multicasting over TCP or UDP), requiring only a small-bandwidth protocol to keep track of the subscribers. It addresses the world-wide uniqueness problem of group-based IP multicast, and closely mimics the dynamics of other broadcast mediums, such as radio and television.

Open Question

How will the pricing schemes work for the sending host if the number of subscribing hosts cannot be determined beforehand? Will it be a "possible viewer" scheme that is currently used for cable TV, or can something more intelligent be proposed using the Counting feature of EXPRESS?