Overlay subgroup communication in large-scale multicast applications☆
Introduction
IP multicast is an efficient one-to-many or many-to-many delivery method which can provide a number of operational advantages for content and network providers by reducing the overall resources consumed to achieve such distribution. A single packet transmitted by the source traverses each link in the multicast distribution tree to all receivers in the multicast group. Due to intensive needs for high bandwidth requirement, large-scale interactive applications such as distributed interactive simulations (DIS), video conferencing tools and multi-player games can benefit from IP multicast. Although the members in such applications join a multicast session for some common goal, abundant content, data type and heterogeneity in members' interests naturally lead to preference heterogeneity within sessions [1], requiring frequent communication within subgroups of members sharing common interests/requirements.
A multicast session shared by all members (referred to as the global multicast session) can be used to support subgroup communication. However, this may lead to inefficiency, i.e., packets are delivered to the entire tree, which results in wasted bandwidth and CPU processing power to transmit and handle unnecessary packets. This is referred to as the exposure problem. The exposure problem can be completely eliminated if data is only forwarded along a tree induced by the members of each subgroup, as required. This can be achieved by creating a new multicast session for each subgroup. However, this requires routers to store multicast forwarding state information for each subgroup, which can cause a significant scalability problem as the number of subgroups increases [2], [3]. Thus, mechanisms to handle preference heterogeneity should consider the both exposure problem and the scalability of multicast forwarding state problem.
Most existing approaches to the preference heterogeneity problem focus on developing clustering frameworks, i.e., given a limited number of multicast sessions, determine how to best cluster multiple subgroups into multicast sessions based on a preference matrix [1] or players' positions in a virtual cell [4].
In the paper, we propose a topology-sensitive subgroup communication (TSC) mechanism to support efficient subgroup communication in large-scale multicast applications. Our TSC mechanism allows members in a subgroup to autonomously build a TSC structure consisting of multiple unicast and scoped multicast connections.
For example, consider a distribution tree for a multicast session, G, in Fig. 1. All end nodes are members of G and the black end nodes are members of a subgroup, S. In our scheme, when a wishes to send packets to other members in S, packets will be delivered as follows: (1) a→b via unicast; (2) b→c via unicast; and, (3) c→{d, e, f} via multicasting with a TTL scope of 2 as shown in Fig. 1. We assume that the multicast tree for G is a bidirectional shared one1.
Note that in the example, the use of unicast can suppress the exposure and the use of scoped multicast can reduce duplicate packets traversing the same link. Our approach does not require the creation of new multicast sessions, which can completely eliminate any additional multicast forwarding state except those of the global session. It tries to minimize the exposure by exploiting spatial locality among members within a given subgroup.
Throughout simulations, we study which environments are advantageous to apply the proposed mechanism and other existing approaches, e.g., global multicast tree or unicast. Simulation results show that under various configurations of density and distribution modes of a subgroup, the sensitivity of our TSC mechanism is small compared to others. This is especially beneficial in the case where information about subgroups is not available - as is likely to be the case in practice.
The paper is organized as follows. We discuss related work and contrast them with our work in Section 2. In Section 3 we discuss how to construct and maintain a TSC structure. In Section 4, we evaluate and compare the proposed TSC mechanism with other schemes in various environments. Section 5 concludes the paper.
Section snippets
Related work
The scalability of state associated with multicast forwarding by routers has been one of the challenges in a wide deployment of IP multicast. Reduction of multicast forwarding state at routers can be achieved through aggregation or elimination of non-branching approaches. In [2], multiple multicast forwarding entries are aggregated if entries have adjacent group address prefixes and matching incoming and outgoing interfaces. The goal of dynamic tunnel multicast [8] and REUNITE [9] is to reduce
TSC mechanism
For each subgroup, a TSC structure is created in a TSC mechanism. The TSC structure serves as a communication channel for members in the given subgroup. In this section, we introduce a TSC structure, and describe how to construct and maintain it.
Performance evaluation
We conduct simulations to study various issues and trade-offs in applying the proposed TSC mechanism in multicast applications. Our main goal is to investigate in which environments it is advantageous to apply the proposed mechanism. For comparison, we examine the performance of the following schemes for subgroup communication.
Global Multicast: This represents a scheme that simply uses the original global multicast group G for subgroup communications.
Unicast-Only: This scheme constructs unicast
Conclusions
In this paper, we designed and evaluated a topology-sensitive subgroup communication mechanism to handle the preference heterogeneity problem in large-scale multicast applications. Our TSC mechanism takes a complete end-to-end approach which eliminates additional creation of multicast groups. Depending on the local density of subgroup members, members in the session self-configure into islands and forwarding structures. Within islands, scoped multicast is used to derive benefit from clustered
Acknowledgements
We would like to thank Daniel Ng for help on simulation in ns-2. This work is supported by the National Science Foundation Grant CNS-0509355.
References (19)
- et al.
An evaluation of preference clustering in large-scale multicast applications
in: Proceedings of IEEE Infocom
(2000) - et al.
On the aggregatability of multicast forwarding state
in: Proceedings of IEEE Infocom
(2000) - et al.
An analysis of multicast forwarding state scalability
in: International Conference on Network Protocols ICNP
(2000) - et al.
Issues in designing a communication architecture for large-scale virtual environments
in: Proceedings of Networked Group Communication Workshop
(1999) - et al.
Protocol Independent Multicast-Sparse Mode (PIM-SM): Protocol Specification, RFC 2362, June
(1998) - et al.
Border Gateway Multicast Protocol (BGMP): Protocol Specification, IETF draft, draft-ietf-bgmp-spec-01.txt, March
(2000) - et al.
Core based trees (CBT): An architecture for scalable inter-domain multicast routing
in: Proceedings of ACM SIGCOMM
(1993) - et al.
Forwarding state reduction for sparse mode multicast communication
in: Proceedings of IEEE Infocom
(1998) - et al.
REUNITE: a recursive unicast approach to multicast
in: Proceedings of IEEE Infocom
(2000)
Cited by (0)
- ☆
Partial work of this paper was presented at proceedings of the 4th international workshop on networked group communication (NGC), Oct. 2002