An Efficient and Scalable Multicast Algorithm That Accommodates Dynamic Groups

Abstract—This paper proposes an efficient and scalable multicast algorithm that accommodates dynamic groups. Our protocol relies on a shared tree architecture to deal with the problems of scalability and group dynamics. Our algorithm is based on the communication model developed by Bhat et al [2] that considers both network and node heterogeneity. Our algorithm uses a modified version Bhat et al. [3] heuristics for multicasting a message to the group.

M

I. INTRODUCTION

any applications such as teleconferencing, distributed games and any collaborative multimedia application require an efficient group communication. Apart from efficiency, group dynamics and group scalability are becoming an important requirement for distributed applications.

Most of the research [2,3] concerning multicast communication has focused on minimizing the time required for a group with a static number of nodes to receive the message. The problem with groups is that very few of them are static, because a static group would mean that all the members join at the same time and leave at the same time which is almost never the case.

SEAM [1], scalable and efficient ATM multicast, is one of the protocols developed to support IP multicasting in ATM networks. This protocol emphasizes on managing efficiently dynamic nodes by using a shared tree architecture. Research on this protocol has proven to be very promising in dealing with scalability issues and the handling of dynamic nodes, but this protocol can only be used in ATM networks and almost no research has been done on the speed of multicasting a message in this group.

Other protocols for multicasting such as PIM-SM [4] (protocol independent multicast – sparse mode) and CBT [5] (Core based trees) are also based on a shared tree structure and have been proven efficient in dealing with scaling issues and dynamic nodes. But the same problem as for SEAM [1] arises for these protocols concerning the delay for multicasting a message to the group.

A source based tree architecture is usually much more efficient for multicasting a message to the group but unfortunately this structure doesn’t scale at all and does not handle well group dynamics. In a group of N members we need to establish N connections at the setup and tear down N connections when we want to leave whereas in shared tree architecture we only handle one connection for the “join” phase and one connection for the “leave” phase. We find ourselves in a dilemma, where “our dynamic group handling performance” is going to limit our speed of multicasting a message to the group and vice versa.

Focusing on minimizing the time required for a group to receive a multicast message, Bhat et al. [2] developed very efficient algorithms for multicasting messages, which are independent from the architecture of the group. Furthermore these algorithms take into account the heterogeneity in both the nodes and the network.

Since Bhat et al [2] algorithms are applicable to both shared and source based trees, we will use a shared tree for the architecture and a modifed version of Bhat’s heuristics for multicasting a message.

The rest of the paper is organized as follows. In section 2, we discuss related works essentially SEAM [1] and Bhat et al [2] communication model and heuristics. Section 3 presents our algorithm. Section 4 concludes our paper and lays the guidelines for future work.

II. RELATED WORKS

A. Scalable and Efficient ATM Multicast

SEAM is multipoint-to-multipoint architecture used for a multicast in ATM networks with multiple senders and receivers [2]. The scheme accommodates well scaling issues and group dynamics.

1) Structure

SEAM uses a single shared spanning tree for all senders and receivers. This choice was made after simulations done on both single shared trees and sources based trees. These simulations evaluated two parameters: the “Latency”, the time it takes to create the state in the network to include the new member in the group [1] and “Resource consumption”, the amount of processing required to include a new member in the multicast group [1].

Simulations show that for both of these metrics, a shared tree is much more effective than a set of source-based trees for a small and large number of members in the multicast group.

Each group has a core that is used as the focal point for routing signaling messages for the group [1].

2) Joining the group

There are two ways for a node to join a group: member initiated join and core initiated