Elsevier

Computer Communications

Volume 32, Issue 2, 12 February 2009, Pages 257-267
Computer Communications

Fast and proximity-aware multi-source overlay multicast under heterogeneous environment

https://doi.org/10.1016/j.comcom.2008.10.007Get rights and content

Abstract

Overlay multicast has been considered as one of the most important developments for the next generation Internet infrastructure. In this paper, we consider overlay multicast in the scenarios where any participant node is a potential data source. Existing multicast algorithms for single-source always require a long time to deliver messages or have high maintenance overhead when multiple data sources are allowed. There are other algorithms that are designed for multi-source scenarios. But they consume too much network resources and have a long convergence time because of proximity ignorance. To address the issues, we present FPCast, which leverages node heterogeneity and proximity information at the same time. Physically close nodes are grouped into clusters and each cluster selects a powerful, stable node as its rendezvous point. The rendezvous nodes form a DHT-based structure. Data messages are replicated and forwarded along implicit, source specific, and heterogeneity-aware multicast trees. We further reduce the delivery delay by introducing probabilistic forwarding scheme. We show the average delivery path length converges to O(logn) automatically (n is the number of nodes in the overlay). The simulation results demonstrate the superiority of our algorithm in terms of message delivery time and network resource consumption, in comparison with the previous randomized algorithms. The algorithm is also resilient to node failures.

Introduction

Overlay multicast has been considered as a promising alternative to the un-widely deployed IP multicast and is one of the most important developments for the next generation Internet. In this paper, we study overlay multicast to support the applications in which there are multiple data sources. The applications include distributed multi-player games, group communication for large-scale systems, teleconferencing, remote collaboration, etc. Since these applications always consist of a large number of dynamic nodes which are heterogeneous in terms of node capacity and spread over the entire Internet, it is a challenge work to design an efficient multi-source overlay multicast algorithm. In our opinion, an efficient multi-source overlay multicast algorithm should meet at least the following basic requirements.

  • Scalable performance. The algorithm should be decentralized and support node churns (i.e. joins, graceful departures and failures). As the system size grows, the efficiency should degrade gracefully.

  • Low delay. Some applications (e.g. group communication) have soft-time constraints while others (e.g. distributed games and teleconferencing) are interactive applications. Lower delay is the main design objective.

  • Proximity-aware. In order to reduce the network resource consumption and enable fast message delivery, the algorithm should take node proximity information into consideration. The proximity of two nodes is measured by the latency in the physical IP topology. The closer two nodes are in the IP topology, the more proximate they are.

  • Heterogeneity awareness. Nodes have different capacities. The algorithm should account for node heterogeneity in terms of capacity so that the load on a node is proportional to its capacity (i.e. achieving a load balance state). Otherwise, some nodes are overloaded while others are underloaded.

  • Reasonable network resource consumption. Both the overlay maintenance overhead and the cost for message delivery should be small.

The single-source overlay multicast algorithms can be classified into two categories: tree-based algorithms [5], [24] and data-driven algorithms [30], [15]. Tree-based algorithms build one tree or several disjoint trees for the specified single-source. The multicast trees that are optimal for one source may be bad for others [7], while building and maintaining one tree for each potential node are too costly. Data-driven algorithms suffer a basic control-overhead-versus latency tradeoff [24]. Thus, they are not suitable for the delay sensitive applications that require fast message delivery, or they bring considerable control-overhead.

In the past, there have been several algorithms for applications with multiple data sources [11], [29], [9], [7]. However, few of them meet above basic requirements. The probabilistic gossip based hybrid push/pull schemes [11] bring lots of duplicate messages and only guarantee probabilistic convergency. Structured P2P network based schemes [9], [29] implement multicast service using implicit trees. However, the overlays are too strict to be optimized and the maintenance overhead is a major design concern [6]. ACOM [7], on the other hand, is based on unstructured overlays. Long delay and high redundancy (i.e. the number of duplicated messages) are its two major concerns. A common and major limitation of these algorithms is proximity ignorance. Therefore, they would consume unnecessary network resources (e.g. backbone bandwidth) and have a long convergence time.

In this paper, we present FPCast, a fast and proximity-aware multi-source overlay multicast algorithm. FPCast is motivated by two facts. First, although the structured overlay networks need considerable maintenance overhead and have relative rigid structures, their predefined structures (e.g. ring and hypercube) give us useful information to extract multicast trees. Second, measurement results in [25], [18] have shown that there exist some relative powerful and stable nodes in large-scale wide area distributed applications.

Therefore, we organize nodes in a two-layer structure: the upper layer is based on DHT protocol and composed of powerful and stable nodes (denoted as Dnodes), while the nodes at the lower layer (denoted as Onodes) attach to physically close Dnodes. Obviously, a Dnode and the Onodes attaching to it constitute a cluster and the Dnode acts as the rendezvous point of that cluster. Data items are replicated and forwarded along implicit, source specific, and heterogeneity-aware multicast trees. We further reduce the delivery delay by introducing probabilistic forwarding scheme in the DHT structure. The analysis results show that the multicast message from any source can been delivered to all the other nodes within O(logn) hops, where n is the number of member nodes. We evaluated the performance of the FPCast algorithm via comprehensive simulations. The results demonstrate the superiority of our algorithm in terms of message delivery time and network resource consumption, in comparison with the previous randomized algorithms. The algorithm is also resilient to node failures.

The rest of the paper is organized as follows: Section 2 provides a survey of related work. Section 3 describes the construction of hierarchical structure in detail. Section 4 gives the details of the basic multicast algorithm, followed by the enhanced algorithm in Section 5. We evaluate our scheme through simulation experiments and show the results in Section 6. Finally, we conclude our work in Section 7.

Section snippets

Related work

Tree-based multicast algorithms [5], [24] designed for single data source are not applicable for the applications with multiple sources. Building and maintaining one tree for each potential node are too costly, while using single tree or fixed small number of trees is not a good choice for that an optimal tree for one source may not be an optimal one for others and the traffic would be concentrated on the tree links [7]. Data-driven multicast algorithms [30], [15] suffer a

Hierarchical structure construction

In this section, we detail the construction of hierarchical structure. Fig. 1 captures the hierarchical overlay in our algorithm. Powerful and stable nodes are organized in a Chord ring, labeled as Dnodes. Ordinary nodes, labeled as Onodes, attach to physically close Dnodes. Although we use Chord [22] as a representative DHT protocol, it is straightforward for other DHT protocols.

Obviously, a Dnode and the Onodes attaching to it constitute a cluster and the Dnode acts as a rendezvous point of

Basic multicast algorithm

An Onode initializes a multicast by transferring the multicast message to its Dnode. When a Dnode y receives a new multicast message from its Onode or initializes a multicast by itself, it delivers the message to other Dnodes through an implicit tree which is based on node finger tables defined and detailed in Chord [22]. The finger table at node y contains the information of O(lognd) Dnodes and the ith entry in the table is 1/2lognd-i of the ring away from y, where nd is the number of Dnodes

Enhanced multicast algorithm

In this section, we aim to enhance the basic multicast algorithm to enable fast delivery.

In the basic multicast algorithm, a Dnode is the root of the implicit tree within the cluster of which it is a rendezvous point. Since the nodes within a cluster are always physically close to each other, the delivery should be fast. Thus, we believe the delivery delay is dominated by the delay of Dnodes. This is confirmed by our simulation result shown in Fig. 3. The simulation set up is described in

Experiment setup

We develop an event-driven simulator to evaluate the performance of our algorithm. We also compare the algorithm with limited-degree flooding (LD-flooding), CAM [29] and ACOM [7]. The metrics we mainly focus on are as follows:

  • M1.

    Average length of delivery path: the average number of overlay hops it takes a message to reach a node.

  • M2.

    Average delivery delay: the average time it takes a message to reach a node.

  • M3.

    Redundancy rate: Let m denote the total number of messages and n denote the number of nodes,

Conclusions

This paper presents FPCast algorithm for multi-source overlay multicast. The algorithm leverages both node heterogeneity and proximity information to enable fast and low-cost message delivery. Nodes form a proximity-aware hierarchical structure. Data messages are distributed along source specific, implicit multicast trees. We prove that the message from any source can be delivered to all other nodes within O(logn) hops, where n is the number of nodes in the overlay. We further reduce the

Acknowledgments

This work is supported by National Basic Research Program of China with grant No. 2007CB310702, by National Natural Science Foundation of China with grant No. 60873242 and No. 90604015. The authors thank the anonymous reviewers for their valuable comments.

References (30)

  • T. Asano et al.

    Space filling curves and their use in geometric data structures

    Theoretical Computer Science

    (1997)
  • S. Banerjee, B. Bhattacharjee, C. Kommareddy, Scalable application layer multicast, in: Proceedings of the ACM Annual...
  • M. Bishop, S. Rao, K. Sripanidkulchai, Considering priority in overlay multicast protocols under heterogeneous...
  • M. Castro, P. Druschel, Y.C. Hu, A. Rowstron, Exploiting network proximity in distributed hash tables, in: Proceedings...
  • M. Castro, P. Druschel, A.-M. Kermarrec, A. Nandi, A. Rowstron, A. Singh, SplitStream: high-bandwidth multicast in...
  • M. Castro, M. Jones, A.-M. Kermarrec, A. Rowstron, M. Theimer, H. Wang, A. Wolman, An evaluation of scalable...
  • S. Chen et al.

    ACOM: Any-source capacity-constrained overlay multicast in non-DHT P2P networks

    IEEE Transaction on Parallel and Distributed Systems

    (2007)
  • Y.-h. Chu et al.

    A case for end system multicast

    IEEE Journal on Selected Areas in Communication (JSAC)

    (2002)
  • S. El-Ansary, L.O. Alima, P. Brand, S. Haridi, Efficient broadcast in structured P2P networks, in: Proceedings of the...
  • K.P. Gummadi, R. Gummadi, S.D. Gribble, S. Ratnasamy, S. Shenker, I. Stoica, The impact of DHT routing geometry on...
  • A.-M. Kermarrec et al.

    Probabilistic reliable dissemination in large-scale systems

    IEEE Transactions on Parallel and Distributed Systems

    (2003)
  • Z. Li, G. Xie, Z. Li, Towards reliable and efficient data dissemination in heterogeneous peer-to-peer systems, in:...
  • Z. Li et al.

    Efficient and scalable consistency maintenance for heterogeneous peer-to-peer systems

    IEEE Transactions on Parallel and Distributed Systems

    (2008)
  • Z. Li, Z. Zhu, G. Xie, Z. Li, Efficient multi-source data dissemination in peer-to-peer networks, in: Proceedings of...
  • V. Pai, K. Kumar, K. Tamilmani, V. Sambamurthy, A.E. Mohr, Chainsaw: eliminating trees from overlay multicast, in:...
  • Cited by (6)

    • Stateless overlay multicast with in-packet bloom filters

      2017, Journal of China Universities of Posts and Telecommunications
    • An overlay multicast protocol for live streaming and delay-guaranteed interactive media

      2012, Journal of Network and Computer Applications
      Citation Excerpt :

      The tree-based P2P approach organizes peers into an explicit data delivery tree. Some P2P approaches also utilize the multi-tree structure for issues like heterogeneity and redundancy (Castro et al., 2003; Li et al., 2009; Venkataraman et al., 2006; Dan et al., 2007). By contrast, the mesh-based P2P approach maintains a set of partners for each peer, and exchanges the data availability information without an explicit structure support.

    • A mobile learning system for supporting heterogeneous clients based on P2P live streaming

      2012, 2012 6th International Conference on Distributed Smart Cameras, ICDSC 2012
    • A subscription overlay network for large-scale and cost-efficient any source multicast

      2011, Conference Proceedings of the IEEE International Performance, Computing, and Communications Conference
    • Proximity-aware DHT for efficient lookup service in peer-to-peer applications

      2011, Proc. - 14th IEEE Int. Conf. on Computational Science and Engineering, CSE 2011 and 11th Int. Symp. on Pervasive Systems, Algorithms, and Networks, I-SPA 2011 and 10th IEEE Int. Conf. on IUCC 2011
    • An administrable and manageable P2P-SIP based VoIP system

      2010, Jisuanji Yanjiu yu Fazhan/Computer Research and Development
    View full text