Aggregated Multicast — A Comparative Study

Jun‐Hong Cui, Jinkyu Kim, Dario Maggiorini, Khaled Boussetta, Mário GerlaLecture Notes in Computer Science 2002 · DOI · OpenAlex

Abstract

Aggregated Multicast (AM) is a networking strategy designed to improve the scalability of IP multicast by allowing multiple multicast groups to share a single distribution tree. This study evaluates AM's ability to reduce router state memory at the cost of minor bandwidth overhead. Simulations using the vBNS backbone topology demonstrate that AM can achieve up to a 70% reduction in forwarding state, offering a viable solution for large-scale multicast deployment.

Introduction: The State Scalability Problem

IP Multicast was designed to save bandwidth by ensuring data packets are only duplicated at "forks" in a network tree, rather than sending individual copies to every receiver. While efficient for bandwidth, it creates a "state explosion" problem. In conventional IP multicast, every router on a delivery tree must maintain a unique forwarding entry for every single active group. As the number of simultaneous multicast groups grows, core routers run out of memory and packet-lookup speeds degrade.

Standard multicast distribution models including Source-Specific and Shared Trees.

Aggregated Multicast addresses this "state scalability" issue by decoupling the multicast group from the delivery tree. Instead of a 1-to-1 mapping, multiple groups are "aggregated" into a single shared tree.

Methods and Implementation

The authors propose and evaluate Aggregated Multicast (AM) using a centralized multicast controller to manage tree assignments. The core of the approach is a Tree Matching Algorithm that considers two types of "imperfect" matches to maximize aggregation:

  • Leaky Match: The tree reaches more nodes than the group needs, resulting in minor bandwidth waste.
  • Incomplete Match: The tree doesn't reach all group members, requiring "tunnelling" (encapsulating packets) to bridge the gaps.
Network topology showing the use of tunnels to support incomplete matches in an aggregated multicast domain.

To evaluate the scheme, the authors used the SENSE Simulator (Simulation Environment for Network System Evolution) on the vBNS Backbone Topology, a real-world network map of 43 nodes. Group memberships were generated using a Random Node-Weighted Model to mimic realistic user locality, with session dynamics modeled via a Poisson process.

Key Insights

  • The "Close Enough" Principle: Perfect tree matches are not required for efficiency. By allowing a "leaky" match, networks can drastically consolidate the number of trees core routers need to track.
  • Transit vs. Terminal State: The study highlights that while state reduction at terminal nodes is difficult, significant gains can be made at "transit nodes" (core routers), where scalability matters most.
  • Tunnelling as a Scalability Lever: The authors found that allowing tunnels for "incomplete matches" had a more significant impact on state reduction than simply allowing bandwidth waste. Tunnelling allows for smaller, more efficient "base" trees to be shared across more groups.
  • Sub-linear Growth: Unlike standard schemes where the number of trees grows linearly with groups, AM's aggregation gain actually improves as more groups are added, as there are more opportunities for sharing.

Key Results

  • State Reduction: Aggregated Multicast achieved up to a 70% reduction in forwarding state at transit nodes compared to conventional schemes.
  • Tree Consolidation: In a simulation of 2,500 concurrent groups, standard Unidirectional Shared Tree (UST) schemes required 2,500 trees. Aggregated Multicast handled the same load with only 1,150 trees under perfect match constraints, and even fewer when allowing for overhead.
  • Efficiency Trade-off: These scalability gains were achieved with manageable overhead, typically sacrificing 10-30% in bandwidth or tunnelling costs to ensure the network remains functional at scale.

Conclusion and Future Work

Aggregated Multicast provides a promising path toward making IP multicast truly scalable for the modern internet. While this study focused on intra-domain environments, future research is needed to address inter-domain scaling and the development of fully distributed routing protocols. The authors are currently moving from simulations to physical testbed validation to evaluate the scheme under real-world traffic conditions.

Summary updated

BibTeX
@inproceedings{cui2002aggregated,
  author = {Jun‐Hong Cui and Jinkyu Kim and Dario Maggiorini and Khaled Boussetta and Mário Gerla},
  title = {{Aggregated Multicast — A Comparative Study}},
  booktitle = {Lecture Notes in Computer Science},
  year = {2002},
  doi = {10.1007/3-540-47906-6\_84},
}
Figure